Mohan P Shell Dissertation

DEVELOPMENT AND APPLICATIONS OF A FLAT TRIANGULAR
ELEMENT FOR THIN LAMINATED SHELLS
By
P. Mohan
A DISSERTATION SUBMITTED TO THE FACULTY OF

VIRGINIA POLYTECHNIC INSTITUTE AND STATE UNIVERSITY
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY
IN
AEROSPACE ENGINEERING
Rakesh K. Kapania, Chairman
Raymond H. Plaut Eric R. Johnson
Efstratios Nikolaidis Owen Hughes
November, 1997
Blacksburg, Virginia
DEVELOPMENT AND APPLICATIONS OF A FLAT TRIANGULAR
ELEMENT FOR THIN LAMINATED SHELLS
by
P. Mohan
Committee Chair: Rakesh K. Kapania
Aerospace Engineering
(ABSTRACT)
Finite element analysis of thin laminated shells using a three-noded flat triangular
shell element is presented. The flat shell element is obtained by combining the Discrete
Kirchhoff Theory (DKT) plate bending element and a membrane element similar to the
Allman element, but derived from the Linear Strain Triangular (LST) element. Though this
combination has been employed in the literature for linear static analysis of laminated plates,
the results presented are not adequate to ascertain that the element would perform well in the
case of static and dynamic analysis of general shells. The element is first thoroughly tested
for linear static analysis of laminated plates and shells and is extended for free vibration,
thermal, and geometrically nonlinear analysis.
The major drawback of the DKT plate bending element is that the transverse
displacement is not explicitly defined within the interior of the element. Hence obtaining the
consistent mass matrix or the derivatives of the transverse displacement that are required for
forming the geometric stiffness matrix is not straight forward. This problem is alleviated by
borrowing shape functions from other similar elements or using simple displacement fields.
In the present research, free vibration analysis is performed both by using a lumped mass
matrix and a so called consistent mass matrix, obtained by borrowing shape functions from an
existing element, in order to compare the performance of the two methods. The geometrically
nonlinear analysis is performed using an updated Lagrangian formulation employing Green
strain and Second Piola-Kirchhoff (PK2) stress measures. A linear displacement field is
used for the transverse displacement in order to compute the derivatives of the transverse
displacement that are required to compute the geometric stiffness or the initial stress matrix.
Several numerical examples are solved to demonstrate the accuracy of the formulation
ii
for both small and large rotation analysis of laminated plates and shells. The results
are compared with those available in the existing literature and those obtained using the
commercial finite element package ABAQUS and are found to be in good agreement. The
element is employed for two main applications involving large flexible structures.
The first application is the control of thermal deformations of a spherical mirror segment,
which is a segment of a multi-segmented primary mirror used in a space telescope. The
feasibility of controlling the surface distortions of the mirror segment due to arbitrary thermal
fields, using discrete and distributed actuators, is studied. This kind of study was required
for the design of a multi-segmented primary mirror of a next generation space telescope.
The second application is the analysis of an inflatable structure, being considered by the
US Army for housing vehicles and personnel. The tent structure is made up of membranes
supported by arches stiffened by internal pressure. The updated Lagrangian formulation of
the flat shell element has been developed primarily for the nonlinear analysis of the tent
structure, since such a structure is expected to undergo large deformations and rotations
under the action of environmental loads like the wind and snow loads. The wind load is
modeled as a nonuniform pressure load and the snow load as lumped concentrated loads.
Since the direction of the pressure load is assumed to be normal to the current configuration
of the structure, it changes as the structure undergoes deformation. This is called the
follower action. As a result, the pressure load is a function of the displacements and hence
contributes to the tangent stiffness matrix in the case of geometrically nonlinear analysis.
The thermal load also contributes to the system tangent stiffness matrix. In the case of
the thermal load, this contribution is similar to the initial stress matrix and hence no
additional effort is required to compute this contribution. In the case of the pressure load, this
contribution (called the pressure stiffness) is in general unsymmetric and its determination
is not straightforward, but can be systematically derived from the principle of virtual work.
The follower effects of the pressure load have been included in the updated Lagrangian
formulation of the flat shell element and have been validated using standard examples in
the literature involving deformation-dependent pressure loads. The element can be used to
obtain the nonlinear response of the tent structure under wind and snow loads.
iii
Acknowledgment
I would like to express my deep sense of gratitude to my advisor, Dr. Rakesh K. Kapania
for providing me with the technical guidance for the entire period of my research.
I would like to thank Dr. Jakubowski for giving me an opportunity to pursue my research
in the area of smart structures.
I would like to thank my committee members, Dr. Plaut, Dr. Johnson, Dr. Nikolaidis
and Dr. Hughes, for reading my thesis and giving valuable suggestions.
I would also like to thank Dr. Plaut, Dr. Johnson, Daniel Hammerand and Jing Li for
the valuable discussions I had with them during the course of my research.
The support provided for my research by NASA and the Army Reaearch Office, through
the grants NAGW-4241 and DAAH04-95-1-0175, is greatly appreciated.
This acknowledgment will not be complete without expressing my gratitude to my
parents, without whose support I could not have been successful in all my endeavors.
iv
Contents
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Literature Survey . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Triangular Membrane Elements . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Triangular Plate Bending Elements . . . . . . . . . . . . . . . . . . . 7
1.2.3 Flat Triangular Shell Elements . . . . . . . . . . . . . . . . . . . . . 11
1.3 Objective and Scope of Present Research . . . . . . . . . . . . . . . . . . . . 15
2 Finite Element Formulation 24

2.1 Membrane Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.2 Plate Bending Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.3 Updated Lagrangian Formulation . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.1 Static analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.3.2 Newton-Raphson method with load control for a concentrated load . 38
2.3.3 Newton-Raphson method with arc-length control for a concentrated load 39
2.3.4 Newton-Raphson method for a thermal load . . . . . . . . . . . . . . 43
2.3.5 Newton-Raphson method for a pressure load . . . . . . . . . . . . . . 44
2.3.6 Dynamic analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3 Numerical examples 51
3.1 Linear Static Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.2 Free Vibration Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Linear Thermal Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4 Geometrically Nonlinear Analysis . . . . . . . . . . . . . . . . . . . . . . . . 58
3.4.1 Static response analysis . . . . . . . . . . . . . . . . . . . . . . . . . 58
v
3.4.2 Dynamic response analysis . . . . . . . . . . . . . . . . . . . . . . . . 60
3.4.3 Thermal postbuckling analysis . . . . . . . . . . . . . . . . . . . . . . 60
3.4.4 Analysis of general shells under deformation-dependent pressure loads 61
4 Shape Control of a Mirror Segment 100

4.1 Introduction and Literature Review . . . . . . . . . . . . . . . . . . . . . . . 100
4.2 Finite Element Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
4.3 Control Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.4 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
5 Analysis of Inflatable Structures 125

5.1 Analysis of a Thin Circular Arch . . . . . . . . . . . . . . . . . . . . . . . . 125
5.2 Analysis of a Tent Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Conclusions 136
Bibliography 141
vi
List of Tables
3.1 Convergence of non-dimensionalized mid-deflection of a simply supported

[0/90/90/0] square plate under doubly sinusoidal load. . . . . . . . . . . . . 63
3.2 Convergence of non-dimensionalized maximum displacements u∗, v∗ and w∗
of a simply supported [-45/45] square plate under a doubly sinusoidal load. 63
3.3 Convergence of the vertical downward deflection at the mid-point of the free
edge of the Scordelis roof. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3
3.4 Convergence of the vertical downward deflection (x10 ) at the load point of a
pinched cylinder. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.5 Convergence of the radial deflection in the direction of the load, at point A,
of a hemispherical shell with a free hole. . . . . . . . . . . . . . . . . . . . . 64
3.6 Convergence of the vertical deflection at the load point A of a hemispherical
shell with a free hole. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.7 Convergence of the radial deflection in the direction of the load, at point A,
of a hemispherical shell, with the vertical deflection constrained along the hole. 65
3.8 Convergence of the downward vertical deflection at the load point A, of a
hemispherical shell, with the vertical displacement constrained along the hole. 65
3.9 Convergence of non-dimensionalized mid- deflection of simply supported
[0/90] and [0/90/90/0] spherical shells under a uniformly distributed load. . 65
3.10 Convergence of first five natural frequencies (Hz) of a square [0/45/-
45/90/90/-45/45/0] cantilever plate using lumped mass. . . . . . . . . . . . 66
3.11 Convergence of first five natural frequencies (Hz) of a square [0/45/-
45/90/90/-45/45/0] cantilever plate using consistent mass. . . . . . . . . . . 66
3.12 Convergence of the first five natural frequencies (Hz) of a [0/22.5] square
cantilever plate using lumped mass. . . . . . . . . . . . . . . . . . . . . . . 67
vii
3.13 Convergence of the first five natural frequencies (Hz) of a [0/22.5] square
cantilever plate using consistent mass. . . . . . . . . . . . . . . . . . . . . . 67
3.14 Convergence of the first five natural frequencies (Hz) of a [152 /0]s skew
cantilever plate (β = 30) using lumped mass. . . . . . . . . . . . . . . . . . 68
3.15 Convergence of the first five natural frequencies (Hz) of a [152 /0]s skew
cantilever plate (β = 30) using consistent mass. . . . . . . . . . . . . . . . . 68
3.16 Convergence of the first five natural frequencies (Hz) of a [0/22.5] skew
cantilever plate (β = 45) using lumped mass. . . . . . . . . . . . . . . . . . 69
3.17 Convergence of the first five natural frequencies (Hz) of a [0/22.5] skew
cantilever plate (β = 45) using consistent mass. . . . . . . . . . . . . . . . . 69

2
3.18 Convergence of first five non-dimensionalized frequencies (ωa ρ/E2 /h) of a
simply supported [0/90/90/0] spherical shell using lumped mass. . . . . . . 70

3.19 Convergence of first five non-dimensionalized frequencies (ωa2 ρ/E2 /h) of a
simply supported [0/90/90/0] spherical shell using consistent mass. . . . . . 71
3.20 Convergence of mid-deflection wmax due to a nonuniform thermal field (Tu =
1, Tl = −1) and maximum in-plane displacements umax , vmax due to a uniform
thermal field (Tu = 1, Tl = 1) for a simply supported [0/90/90/0] square plate. 72
3.21 Convergence of mid-deflection wmax and maximum in-plane displacements
umax , vmax for a simply supported [0/90/0/90] square plate (BC1) under a
non-uniform thermal field (Tu = 1, Tl = 0). . . . . . . . . . . . . . . . . . . 73
umax , vmax for a simply supported [15/-15/15/-15] square plate (BC2) under
a non-uniform thermal field (Tu = 1, Tl = 0) . . . . . . . . . . . . . . . . . . 74
umax , vmax for a simply supported [0/90/0/90] spherical shell (BC1) under
umax , vmax for a simply supported [45/-45/45/-45] spherical shell (BC1) under
4.1 Convergence of non-dimensionalized mid-deflection of a graphite/epoxy

laminate excited by a uniform electric field. . . . . . . . . . . . . . . . . . . 114
viii
4.2 Transverse deflection (x107 ) in m of a piezoelectric bimorph cantilever beam
under uniform electric field, at different locations. . . . . . . . . . . . . . . 114
4.3 Deflection in μm, at point A, of the L-frame . . . . . . . . . . . . . . . . . . 115
4.4 Properties and geometry of the mirror, piezoelectric strip and force actuator. 115
4.5 Temperature distribution at the lower surface of the mirror. . . . . . . . . . 116
4.6 Uncorrected rms and maximum absolute values of the transverse deformations
of Model-1, in μm, due to different temperature distributions (ΔTz =
0.2o C, ΔTxy = 0.5o C). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
4.7 Corrected rms and maximum absolute residual values of the transverse
deformations of Model-1, in μm, due to different temperature distributions,
using 30 piezoelectric strips (ΔTz = 0.2o C, ΔTxy = 0.5o C). . . . . . . . . . . 117
4.8 Corrected rms and maximum absolute residual values of the transverse
deformations of of Model-1, in μm, due to different temperature distributions,
using 121 piezoelectric strips (ΔTz = 0.2o C, ΔTxy = 0.5o C). . . . . . . . . . 117
4.9 Uncorrected and corrected rms values of the transverse deformations of
Model-1, in μm, due to different temperature distributions, using 121 evenly
distributed strips for ΔTz =0.025o C and ΔTxy =0.4o C. . . . . . . . . . . . . 117
4.10 Uncorrected and corrected rms and maximum absolute values of the
transverse deformations of Model-2, in μm, due to different temperature
distributions, using 61 force actuators (ΔTz = 0.2o C, ΔTxy = 0.5o C). . . . . 118
4.11 Uncorrected and corrected rms values of the transverse deformations of
Model-2, in μm, due to different temperature distributions, using 61 force
actuators (ΔTz = 0.025o C, ΔTxy = 0.4o C). . . . . . . . . . . . . . . . . . . . 119
ix
List of Figures
1.1 Tent structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.1 Positive directions of βx and βy . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.1 Geometry of the Scordelis-Lo roof. . . . . . . . . . . . . . . . . . . . . . . . 77

3.2 Geometry of the pinched cylinder . . . . . . . . . . . . . . . . . . . . . . . . 78
3.3 Geometry of the hemispherical shell . . . . . . . . . . . . . . . . . . . . . . 79
3.4 Geometry of the skew cantilever plate. . . . . . . . . . . . . . . . . . . . . . 80
3.5 Normalized modes 2,3,4 and 5 of a [0/45/-45/90/90/-45/45/0] cantilever plate
obtained using consistent mass matrix . . . . . . . . . . . . . . . . . . . . . 81
3.6 Normalized modes 2,3,4 and 5 of a [0/22.5] cantilever plate obtained using
consistent mass matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
3.7 Normalized modes 2,3,4 and 5 of a [152 /0]s skew cantilever plate (β = 30)
3.8 Normalized modes 2,3,4 and 5 of a [0/22.5] skew cantilever plate (β = 45)
3.9 Normalized modes 2,3,4 and 5 of a [0/90/90/0] simply supported spherical
shell obtained using consistent mass matrix . . . . . . . . . . . . . . . . . . 85
3.10 Geometry of the hinged cylindrical panel. . . . . . . . . . . . . . . . . . . . 86
3.11 Snap-through behavior of a [45/-45] hinged cylindrical panel under a
concentrated mid-load. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
3.12 Cantilever beam under a tip moment. . . . . . . . . . . . . . . . . . . . . . 88
3.13 Clamped cylindrical shell under pinching loads (quarter shell model, 30x30
mesh). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
3.14 Cylindrical shell with free ends under stretching loads (1/8th model, 20x20
mesh). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
x
3.15 Hemispherical shell under pinching and stretching loads (u at point a, quarter
shell model, 40x40 mesh). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
3.16 Hemispherical shell under pinching and stretching loads (w at point a, quarter
shell model, 40x40 mesh). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.17 Ring plate under line load at the free edge (full plate, 10x80 mesh). . . . . . 93
3.18 Dynamic analysis of an anisotropic ([45/-45]) hinged cylindrical panel under
a concentrated load at the center. . . . . . . . . . . . . . . . . . . . . . . . 94
3.19 Thermal postbuckling analysis of a simply supported square plate under
uniform temperature rise (full plate, 16x16 mesh). . . . . . . . . . . . . . . 95
3.20 Thermal postbuckling analysis of a clamped cylindrical panel under uniform
temperature rise. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.21 Cantilever beam under uniform external pressure. . . . . . . . . . . . . . . 97
3.22 Cantilever beam under uniform external pressure (comparison of results
obtained with and without pressure stiffness). . . . . . . . . . . . . . . . . . 98
3.23 Thin circular ring under nonuniform pressure (half-ring model, 1x40 mesh, 80
elements). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.1 Geometry and material properties of the L-frame . . . . . . . . . . . . . . . 119

4.2 Finite element mesh for Model-1. . . . . . . . . . . . . . . . . . . . . . . . . 120
4.3 Piezoelectric strip configurations. . . . . . . . . . . . . . . . . . . . . . . . . 121
4.4 Optimal locations of 30 strips. . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.5 Optimal locations of 121 strips. . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.6 Locations of 61 force actuators. . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.1 Convergence of the deflection at point A, of a thin circular arch under a

concentrated apex load P . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.2 Deflection at point A, of a thin circular arch under concentrated apex load P 130
5.3 Convergence of the deflection at point A, of a thin circular arch under a
symmetric nonuniform external pressure . . . . . . . . . . . . . . . . . . . . 131
5.4 Deflection at point A, of a thin circular arch under symmetric non-uniform
external pressure (quarter arch model, 25x65 mesh, 3250 elements, 10296
d.o.f.) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
xi
5.5 Deflection at point A, of a thin circular arch under unsymmetric nonuniform
pressure (half arch model, 25x150 mesh, 7500 elements, 23556 dof) . . . . . 133
5.6 Analysis of a tent structure under a wind load . . . . . . . . . . . . . . . . . 134
5.7 Analysis of a tent structure under a snow load . . . . . . . . . . . . . . . . 135
xii
Chapter 1
Introduction
1.1 Background
The two most widely adopted approaches in the finite element analysis of general shells are
1) use of curved elements based on a suitable shell theory and 2) approximation of the curved
structure by an assemblage of flat shell elements.
The first application of the finite element method to the analysis of shells consisted
of replacing the curved shape with an assemblage of flat elements obtained by combining
independent plate bending and membrane formulations. The membrane-bending coupling
was brought about as a result of anisotropic material properties and the transformation
of the element stiffness matrix evaluated in a local element coordinate system to a global
Cartesian coordinate system. This was soon followed by attempts to develop curved elements
which give a better approximation to the actual shell curvature. However some difficulties
exist in deriving the force-displacement relationships of curved shell elements. Some of the
problems associated with the development of curved shell elements are discussed by Meek
and Tan [1] and the references therein. These difficulties can be eliminated, at the expense of
introducing an approximation to the curved geometry, using an assemblage of flat elements.
Intuitively, as the size of the flat subdivision decreases it would seem that convergence to
the correct solution should occur, and indeed experience shows that such a convergence can
be obtained.
Flat shell elements are more attractive than the curved shell elements as they can be
readily built by combining existing plate and membrane elements. The tedious task of
1
CHAPTER 1. INTRODUCTION 2
computing the element curvatures as required in the case of a curved shell element is
completely avoided in a flat shell formulation. The formulation is extremely simple and
easy to code. The cost of the stiffness formulation for a flat shell element is expected to be
much less than that for a curved shell element. Hence flat shell elements are more suitable
for nonlinear analysis, as the stiffness matrix will have to be evaluated and factorized several
times during a nonlinear analysis. The choice of the membrane and plate bending elements
that are combined to obtain a flat shell element depends on many factors like in-plane
rotational singularity, shear locking and spurious zero-energy modes.
In-plane rotational singularity occurs in the case of a flat shell element when the
membrane formulation does not include an in-plane rotational degree of freedom or the
drilling degree of freedom. The degrees of freedom normally found in a plate bending
element are a transverse displacement and out-of-plane rotations. A flat shell element
formed by combining a plate bending element and a membrane element which has only in-
plane displacements as degrees of freedoms can still exhibit an in-plane rotational degree of
freedom at a node in a global Cartesian co-ordinate system. This occurs when all the elements
sharing the node are not coplanar and the local coordinate system of the elements does not
coincide with the global Cartesian co-ordinate system. This is due to the transformation of
the element stiffness matrix computed in a local element co-ordinate system to the global
Cartesian co-ordinate system. If all the elements sharing a node are coplanar and the local
co-ordinate system of the elements coincides with the global Cartesian co-ordinate system,
the absence of in-plane rotational degree of freedom will result in a singular stiffness matrix.
This problem can be alleviated by associating a fictitious stiffness to the in-plane rotational
degree of freedom of the element, but there is no physical reasoning for the use of such
fictitious stiffness.
Shear locking is a phenomenon attributed to the predominance of the transverse shear
energy over the bending energy. Due to the presence of a parameter in the transverse
shear strain energy that is inversely proportional to the thickness of the element, the
transverse shear energy becomes very large compared to the bending energy as the thickness
of the element becomes very small and hence results in an overly stiff solution. In other
words the bending energy becomes very small or negligible compared to the shear energy,
resulting in spurious modes and incorrect results. The shear locking problem can be
avoided by completely neglecting the shear strain energy and applying discrete Kirchhoff
constraints. This has the effect of relating the shell normal rotations in terms of the transverse
displacement and its derivatives (or slopes) at discrete locations (usually at the corner and
mid-side nodes) and thus eliminating the shell normal rotations at these locations. In order
that the element can be used for both thick and thin shells, a shear stiffness multiplication
factor can be used such that a proper balance between the bending and transverse shear
energy is maintained in both thick and thin regimes. The shear locking can also be avoided by
using a reduced integration scheme to compute the stiffness corresponding to the transverse
shear energy, since such a scheme results in a softer element. As mentioned above, the
major drawback of using reduced integration is the creation of spurious zero-energy modes
which will have to be removed using some stabilization techniques, resulting in a complex
formulation.
An element without any of the above mentioned problems is certainly desirable, as these
problems have dramatic influence on the correctness of the results, and also the formulation
could become complicated due to the additional effort required to alleviate these problems.
Membrane and plate bending elements without any of the above mentioned problems have
been successfully developed and hence they can be combined to obtain a reliable flat shell
element. Due to the extreme simplicity of both the membrane and plate bending formulations
and the availability of high speed parallel computers, the element development cost and the
computational cost of the analysis are expected to be less than those of curved shell elements,
and hence flat shell elements seem to be promising candidates for economic and yet accurate
analysis of large structures, especially for nonlinear analysis.
The literature on membrane and plate bending elements is so vast that any attempt
to provide a comprehensive review will be futile. Only those elements that are of historic
importance and those that have been widely used in the existing literature are discussed in
the following sections. Further, the discussion is limited to triangular elements.
1.2 Literature Survey
1.2.1 Triangular Membrane Elements

The CST (Constant Strain Triangle) and the (LST) Linear Strain Triangle [2] have been
predominantly used with suitable plate bending elements. The CST element is the simplest
membrane element one could possibly think of. It has three corner nodes with two in-plane
displacements per node and the variation of the dispacements within the element is linear.
The LST element has three corner and three mid-side nodes with two in-plane displacements
per node and the displacements vary quadratically within the element. Though it is well
known that the performance of the LST element is superior to that of the CST element, the
later has been used in the majority of the flat shell elements that are found in the existing
literature, since the nodal degrees of freedom of the CST element can be easily matched
with many three node plate bending elements. The drawback of both the CST and the LST
elements is the absence of a drilling degree of freedom, and hence they suffer from in-plane
rotational singularity as mentioned in the previous section.
The first successful triangular membrane element with drilling degrees of freedom was
introduced by Allman [3]. This element, popularly known as the Allman triangle can, be
considered as a subset of a more general formulation later proposed by Allman [4]. In this
later formulation the in-plane displacements are represented by a cubic polynomial in area
coordinates. If the third order terms are dropped, this element reduces to the well known
Allman triangle. The latter formulation [4] is not so popular, probably due to the complexity
of the interpolation functions. The formulation of the Allman triangle is extremely simple,
but it suffers from a peculiar singularity which occurs when all the nodal displacements are
zero and the nodal rotations are identical. In most of the problems encountered in structural
mechanics, this singularity is automatically removed due to the application of kinematic
boundary conditions, unlike complex stabilization schemes that are required to remove the
spurious zero energy modes that are created due to the use of reduced order integration.
Cook [5] showed that the Allman triangle can be derived from the LST element using
a simple transformation which relates the mid-side displacements in terms of the corner
displacements and rotations. Since the stiffness matrix of the LST element is available in
many standard textbooks, it is easier to obtain the stiffness matrix of the Allman element
from that of the LST element using the transformation of Cook [5] than to obtain the stiffness
matrix from the shape functions originally proposed by Allman [3].
Another successful triangular membrane element with vertex rotations was developed
by Bergan and Felippa [6] using the free formulation originally suggested by Bergan and
Nygard [7]. In the free-formulation, the element stiffness matrix is derived by expressing
the requirements of the standard patch test in terms of a set of constraints imposed on the
element stiffness matrix, without the need for a variational principle. The patch test is a
check which ascertains whether a patch of elements subject to a state of constant strain
reproduces exactly the constitutive behavior of the material and results in correct stresses
(Zienkiewicz and Taylor [8]). If it does, then the finite element model represents the real
material behavior and in the limit, as the size of the elements decreases, would reproduce
exactly the behavior of the real structure (Zienkiewicz and Taylor [8]). The patch test was
originally proposed in the historic work by Bazeley et al. [9], which is discussed in the next
section.
Carpenter et al. [10] presented a membrane element with degrees of freedom and
displacement field identical to that of the Allman triangle [3], but they used a one-point
integration scheme to compute the element stiffness matrix. As a result of the reduced
order integration, the element exhibits spurious zero energy modes which were removed by
Fish and Belytschko [11] using a stabilization scheme. This was done by adding terms to
the strain-displacement relations of the one-point integrated Allman element such that the
spurious modes were not activated.
Cook [12] presented a more straightforward stabilization scheme, in which two matrices
are added to the stiffness matrix of the one-point integrated Allman element. The first,
denoted as [k1], was originally proposed by Zienkiewicz [13] to suppress the in-plane
rotational singularity. The second, denoted as [k2 ], is derived based on the idea originally
proposed by MacNeal and Harder [14]. The physical basis on which these two matrices are
derived and how the spurious modes are suppressed by these matrices is not clearly explained
in the existing literature. The membrane element stabilized using the two matrices [k1 ] and
[k2 ] was denoted by Cook [12] as T1.
Cook [12] also presented membrane elements obtained by subdividing T1 into 3, 4 and
16 elements denoted as T2, T3 and T4, respectively. He assessed the performance of these
four elements for a cantilever beam under in-plane forces. He compared the deflections and
stresses obtained from the finite element model with those obtained using beam theory. The
finite element mesh used (both regular and irregular) was very coarse with only 8 elements.
The elements with subdivisions did not show marked improvement over T1 in the case of
the regular mesh, whereas there was some improvement in the case of the irregular mesh.
This is not surprising since the aspect ratio of the elements used in the irregular mesh
greatly improves when subdivided. The results presented do not show any superiority of the
elements T2, T3 and T4, as such an irregular mesh with elements of very poor aspect ratio
may not be used if accurate results will have to be obtained. Cook [12] did not compare
the performance of the new elements with that of the Allman triangle, especially for finer
meshes. He also remarked that the cost of generating the elements with subdivisions is at
least 2.5 times that of T1 and expressed concern regarding the computational expense of the
elements T2, T3 and T4.
The element T1 might be a better choice than the Allman triangle since it is softer and
not computationally very expensive, but a drawback of the stabilization scheme proposed by
Cook [12] is the use of arbitrary multiplication factors which are problem dependent. Values
for these factors were selected by Cook [12] based on how accurately the displacements at
certain nodes were predicted. Such a trial and error process of determining the factors may
be very expensive when the problem size is very large and might be less economical than
using the Allman triangle [3] with a finer mesh.
Cook [15] presented derivatives of the CST element, obtained by adding supplementary
stiffness matrices such that the resulting element has vertex rotations. He presented seven
elements named E1 to E7, out of which E3 is the Allman triangle and E7 is the CST element.
As in the case of the formulations presented earlier by Cook [12], the new elements required
three arbitrary parameters, except of course the elements E3 and E7. The results presented
by Cook [15] for a swept panel using a fine mesh do not indicate any marked improvement
of performance of the new elements over the Allman triangle (E3). He found that when
the element E2 was combined with the well known Discrete Kirchhoff Theory (DKT) plate
bending element of Batoz et al. [16], for certain values of the parameters, better results were
obtained for plane problems whereas very poor results were obtained for many standard shell
tests (Scordelis-Lo roof, Pinched cylinder and Hemispherical shell). Cook [15] also found that
the elements E1-E6 gave overly stiff results in the case of the hemispherical shell. Cook [15]
came to a conclusion that the goal of developing a flat triangular element that works well for
both plane and shell problems had not been achieved. This is probably an understatement
since successful flat triangular shell elements had been developed by then. A review of flat
triangular shell elements is presented later in this Chapter.
The above mentioned membrane formulations are all based on displacement models.
Cook [17] presented a plane triangular element with rotational d.o.f., formulated using the
assumed-stress hybrid method, which is based on independent prescriptions of stresses within
the element and displacements on the element boundary. Membrane elements based on a
hybrid formulation have also been presented by Allman [18], Yunus et al. [19] and Cazzani
and Atluri [20]. Hughes and Brezzi [21] investigated variational principles that permit
independent interpolations for the displacement and rotational fields. Membrane elements
based on this principle were presented by Ibrahimbegovic [22] and Iura and Atluri [23]. In a
three part series, Alvin et al. [24], Felippa and Militello [25] and Fellipa and Alexander [26]
presented high performance membrane triangular elements with vertex rotations formulated
using advanced element construction techniques, namely, Extended Free Formulation (EFF)
and Assumed Natural Deviatoric Strain (ANDES) formulation.
Some of the above mentioned membrane elements are known to give more accurate results
than the Allman element even for coarse meshes. The same accuracy may be obtained using
the Allman element by using a finer mesh without much demand on the computational time
due to the simplicity of the formulation. This is probably the main reason for the popularity
of the Allman element in spite of the availability of other membrane elements which can
outperform the Allman element.
1.2.2 Triangular Plate Bending Elements

The choice for a three-noded plate bending element is not as limited as is the case for a three-
noded membrane element. Early attempts to develop three-noded plate bending elements
with 9 degrees of freedom (transverse displacement and two out-of-plane rotations per node)
using interpolation functions in Cartesian coordinates were unsuccessful. For example,
combining the terms x2y and xy 2 results in a singular coefficient matrix for some orientations
of the element. In addition to this problem, the elements derived using such displacement
functions do not satisfy the continuity of the normal slope along inter-element boundaries.
This non-conformity was believed to be the stumbling block in the development of triangular
plate bending elements until the introduction of the first successful non-conforming element
by Bazeley et al. [9]. They used the area coordinates instead of the Cartesian coordinates
and derived cubic displacement functions in terms of the nodal transverse displacements and
normal rotations. The element was non-conforming since the normal slope along the element
edges is parabolic, which cannot be uniquely represented by the corner connections. They
derived a conforming element by adding certain rational functions to the shape functions
developed for the non-conforming element such that normal rotations along the edges now
varied linearly, thereby fulfilling the continuity of the normal rotations. The main drawback
of this conforming element is the requirement of a 16 point numerical integration scheme.

Bazeley et al. [9] found that even for fairly coarse meshes the non-conforming type
solution was superior to that of the conforming type solution. They also found that for
some mesh configurations the non-conforming element did not show convergence to the
exact solution. They came to the conclusion that a necessary and sufficient condition for the
convergence of a non-conforming element is that the element should be capable of exhibiting
a state of constant strain in the limit as the element size is made infinitesimal. This has
evolved into what is called the “patch test”, which can be interpreted as a check to ascertain
whether a patch of elements subject to a constant strain reproduces exactly the constitutive
behavior of the material and results in correct stresses. Though this test would have to be
passed when the size of the patch is infinitesimal, the standard practice is to apply the test
to a patch of elements of finite size. The patch test is administered in several forms. For
details the reader is directed to Ref. [8] and the references therein.
Morley [27] presented a non-conforming triangular element with a quadratically varying
displacement field expressed in terms of the corner deflections and the normal slopes at the
mid-sides. Morley found that this element gave satisfactory values of bending moments, but
gave overly soft values for the deflections even for finer meshes. Several attempts have been
made to develop successful conforming and non-conforming triangular elements. Some of
them have been summarized by Yang [2], with a brief description of each formulation. A
catalog of triangular plate bending elements developed prior to 1984 is given by Hrabok and
Hrudey [28].
Batoz et al. [16] studied several three-noded triangular plate bending elements with 9
degrees-of-freedom, two rotations and a translation per node. They came to a conclusion
that the Discrete Kirchhoff Theory (DKT) element based on the displacement formulation
and the Hybrid Stress Model (HSM) element are the most efficient and reliable three-node
plate bending elements for the analysis of thin plates.
In the case of the DKT element the shear strain energy is completely neglected, but
the formulation proceeds first by assuming that the shear deformations are present and the
strain energy is expressed in terms of the normal rotations. Due to the presence of only the
first derivatives of the normal rotations in the expression for the strain energy, the shape
functions used to describe the normal rotations have to satisfy only the Co continuity. The
normal rotations are first expressed in terms of the nodal values at the corner and mid-
side nodes using the standard quadratic shape functions in area coordinates. The normal
rotations at the corner and mid-side nodes are then eliminated by applying the discrete
Kirchhoff constraints. Though the element is non-conforming, it passes the patch test and
hence converges to the exact solution as the mesh is refined.
The HSM element is based on the hybrid stress or mixed method which was used to
overcome the difficulties that are encountered in displacement models due to the conformity
requirements. The HSM element has a linear distribution of bending moment in the interior,
a cubic displacement variation along the edges and a linear normal slope variation along
the edges. The nodal degrees of freedom of both the DKT and HSM elements are the
same. The DKT element has gained popularity over the HSM element since the formulation
of the HSM element is cumbersome, involving more complicated algebraic manipulations.
Alternative explicit formulation for the DKT element and the FORTRAN code for computing
the stiffness matrix are given by Jeyachandrabose and Kirkhope [29].
Subsequent to the historic work by Batoz et al. [16] there has been increased interest in
developing more efficient triangular plate bending elements. Jeyachandrabose and Kirkhope
[30] constructed an efficient three-noded triangular plate bending element which closely
resembles the DKT element presented by Batoz et al. [16]. In the DKT formulation of Batoz
et al. [16] the Kirchhoff hypothesis is introduced only on the corner and mid-side nodes of the
element. In the formulation by Jeyachandrabose and Kirkhope [30] the Kirchhoff hypothesis
is also introduced at the centroid of the element. This required an approximation for the
normal rotations using seven shape functions. Since the transverse displacement w is not
uniquely defined in the interior of the element proposed by Jeyachandrabose and Kirkhope
[30], as is the case with the DKT element [16], calculation of the derivatives of w at the
centroid is carried out by borrowing shape functions from existing elements. In order to
soften the element, two different smoothing techniques are used, giving two new elements.
FORTRAN listings are also given by Jeyachandrabose and Kirkhope [30] for computing the
element stiffness matrices.
Tessler and Hughes [31] presented a three-noded non-conforming shear deformable
element called MIN3, based on the Reissner-Mindlin theory. In the formulation of the MIN3
element the transverse displacements are expressed in terms of the standard quadratic basis
in area coordinates. The shell normal rotations are assumed to vary linearly over the element.
The transverse displacements at the mid-side nodes are eliminated by imposing a constraint
that the shear strain is continuous along the element edges. This kind of anisoparametric
interpolation and the use of a suitable shear correction factor results in an element which has
the correct rank and does not suffer from shear locking even when full numerical integration
is used. The element is applicable for both thick and thin plates.
Meek and Tan [32] presented a discrete Kirchhoff triangular element with loof nodes (6
Gauss points, 2 on each side). In their plate bending element the transverse displacement is
expressed in terms of the standard quadratic basis in area coordinates using the corner and
mid-side nodal values. The rotations of the shell normal βx and βy about a local Cartesian
coordinates y and x, respectively, are interpolated in terms of the values at the loof nodes.
However, the normal rotations are not expressed using the standard quadratic basis in area
coordinates, due to the existence of a linear combination of the basis called the neutral
function which is zero at all the loof nodes. The existence of such neutral functions in
rectangular elements had been observed by Gopalacharyulu [33] and Irons [34] and they
suggested that the singularity can be eliminated by expanding the basis by adding an extra
node. For a rectangular element with loof nodes, Irons [34] suggested that the centroid be
used as the extra node and also presented an expanded basis. Meek and Tan [32] used an
expanded basis by adding a cubic term to the standard quadratic basis in area coordinates.
They also chose the centroid as the additional node. The normal rotations (βx, βy ) at the
loof nodes and the centroid are eliminated in terms of the rotations of the shell normal
about the edges at the loof nodes by imposing Kirchhoff assumptions at discrete points.
Due to the presence of fourth order terms, a seven point Gauss quadrature rule is required
to numerically integrate the element stiffness matrix of their element. As a result, nonlinear
analysis of large structures using their element might be computationally very expensive.
Though their element passes the patch test, it was found to exhibit good convergence only
in the case of uniformly distributed loads and poor performance under concentrated loads.
Dhatt et al. [35] presented a six node discrete Kirchhoff element called DKTP whose
formulation is similar to that of the well known DKT element. Though the interpolations
used for the transverse displacement and the normal rotations in the DKTP element are one
order higher than those used in the DKT element, the accuracy of the DKTP element was
found to be of the same order as that of the DKT element. Dhatt et al [35] have mentioned
that this is because in both formulations the rotation of the normal to the undeformed
mid-surface of the element about the element edges is assumed to vary linearly along the
edges.
Felippa and Bergan [36] presented a triangular plate bending element based on the free
formulation. They modified the cubic terms of the nine term polynomial in area coordinates
given by Bazeley et al. [9] so that the resulting element passes the patch test. Since the
free formulation is based on the underlining principles of the patch test, elements derived
using this principle are guaranteed to pass the patch test and do not have to undergo the
patch test. Specht [37] presented another modified version of the basis first used by Bazeley
et al. [9] by imposing the condition that the integrated jumps of the displacement and its
normal derivative along the element boundaries is zero. This is also called the interpolation
test for plate bending problems. The modified version has fourth order terms in place of the
cubic terms given by Bazeley et al. [9]. These shape functions in explicit form are given by
Zienkiewicz and Taylor [8] and hence can be readily incorporated into a computer code.
1.2.3 Flat Triangular Shell Elements

No one particular plate bending or membrane element has been claimed to be the best
for the analysis of general shells. The CST, LST, Allman triangle, DKT, MIN3 and the
Morley’s triangle have been widely used in the flat shell formulations that are available in the
existing literature. Flat shell elements based on mixed formulations and other unconventional
methods like the natural mode method and free-formulation have also been successfully
developed. The majority of the flat shell formulations available in the literature have been
developed primarily for non-linear analysis since a non-linear analysis using flat shell elements
is expected to be more economical than that using curved shell elements. The updated
Lagrangian formulation (ULF) has been predominantly used in the non-linear flat shell
formulations available in the existing literature.
In an updated Lagrangian formulation all the variables are referred to a known
configuration, the reference configuration, which is continuously updated during the solution
process. During the deformation process a large portion of the total deformation of the
structure is made up of rigid body motions. If the rigid body modes are removed from the
total displacements, the resulting deformational translations and rotations are very small
and hence a linearized incremental formulation [39] can be used. In a linearized incremental
formulation the stresses are computed using linear strain-displacement relations. The tangent
stiffness matrix contains only the linear stiffness matrix and the initial stress matrix. The
stiffness matrices resulting from the nonlinear terms in the strain-displacement relations are
neglected, resulting in a very economical analysis. If the rigid body modes are not removed all
nonlinear terms in the strain displacement relations will have to be considered for computing
the stresses and the tangent stiffness matrix.
A three node flat triangular shell element was introduced by Argyris et al. [40] for non-
linear elastic stability problems. This formulation, denoted as the “natural mode technique”,
is based on decomposing the displacements into rigid body and straining modes, which are
naturally exhibited by any structure undergoing deformation.
Horrigmoe and Bergan [41] presented flat triangular and quadrilateral shell elements
based on a hybrid stress model for static analysis of isotropic plates and shells. They
employed a linearized updated Lagrangian formulation in which the rigid body modes are
removed from the total displacements. The initial stress matrix was obtained using a lower
order interpolation than that used for the linear stiffness matrix.
Chen [42] presented a three node flat triangular shell element obtained by combining
CST and Morley’s plate bending element for elasto-plastic analysis of isotropic plates and
shells. Chen used the Von Karman theory of large deflection and adopted the Lagrangian
description to describe the deformation of the element. Chen presented numerical examples
that involved only small rotations. Peric and Owen [43] presented a thin shell formulation
identical to that of Chen [42] for elasto-plastic analysis of isotropic plates and shells, but
presented numerical examples involving large rotations.
Bathe and Ho [44] presented a three node flat triangular shell element for static analysis
of isotropic plates and shells by combining the DKT [16] and CST elements. They used
a linearized updated Lagrangian formulation in which the undeformed coordinates are
used to obtain the strain-displacement matrices for all times. The membrane stresses
and internal forces are computed from total displacements as was done by Horrigmoe
and Bergen [41], whereas the curvatures and the bending moments are computed from
incremental displacements and updated at the end of each increment. Bathe and Ho also
used a linear displacement field to compute the initial stress matrix. Due to the use of a
linearized incremental formulation without removing the rigid body rotations from the total
or incremental displacements, their formulation is restricted to small incremental rotations.
The above mentioned restriction in the formulation of Bathe and Ho [44] was overcome by
Hsiao [45] by removing the rigid body modes from the total displacements, in two stages. In
the first stage the undeformed element is rotated to an intermediate position by applying an
out-of-plane rotation about a fixed axis. The element is transformed to its final configuration
by applying an in-plane rotation to the intermediate configuration. Both the membrane
and bending internal force vectors and the membrane stresses are obtained from the total
deformational translations and rotations.
Peng and Crisfield [46] also presented a three node flat triangular shell element identical
to that of Chen [42] , but they removed the rigid body modes from the total displacements
using a procedure similar to that used by Hsiao [45], thereby enabling the use of very large
load increments.
Meek and Tan [47] presented a flat triangular shell element for static analysis of isotropic
plates and shells by combining the LST element and the discrete Kirchhoff plate bending
element with loof nodes [32]. Due to the presence of fourth order terms in the plate bending
formulation, a seven point Gauss quadrature rule is required to numerically integrate the
element stiffness matrix of their element. As a result, nonlinear analysis of large structures
using their element might be computationally very expensive.
Recently Poulsen and Damkilde [48] presented a flat triangular shell element by combining
the LST element and a plate bending element with loof nodes similar to that of Meek and
Tan [32]. The two major differences between the plate bending elements of Meek and Tan
[32] and Poulsen and Damkilde [48] are 1) the seventh term in the expanded basis used
by Meek and Tan [32] is unsymmetric in area coordinates, whereas that used by Poulsen
and Damkilde [48] preserves symmetry, 2) the basis used by Poulsen and Damkilde [48] to
express the transverse displacement has a term ξ1 ξ2 ξ3 in area coordinates called the bubble
function, in addition to the standard quadratic basis, with the centroid as the additional
node. Such a basis was expected to give better load distribution and improved results for
nonlinear analysis. The final degrees of freedom are the same in both the formulations and
the condensation was done by Poulsen and Damkilde [48] as was done by Meek and Tan
[32]. Both Meek and Tan [47] and Poulsen and Damkilde [48] have not presented any shell
problems involving large rotations to demonstrate the performance of the formulation for
such problems.
Fafard et al. [49] presented a six node flat triangular shell element called DLTP for static
analysis of isotropic plates and shells. Their flat shell element is obtained by combining
the six node discrete Kirchhoff theory plate bending element (DKTP) of Dhatt et al. [35]
and the well known LST element. Fafard et al. [49] presented two updated Lagrangian
formulations. The first, termed ULF1, is a total Lagrangian formulation within each step,
with the configuration at the beginning of the step taken as the reference configuration for
the next step. The second, termed ULF2, is a fully updated Lagrangian formulation with the
estimated configuration at the end of each iteration taken as the reference configuration for
the next iteration. In ULF1, all nonlinearities are considered in the computation of stresses
and the tangent stiffness matrix. In ULF2 a linearized incremental formulation is used but
the stresses are computed using nonlinear strain-displacement relations as in ULF1. Fafard
et al. [49] studied the importance of using lower order interpolations for the transverse
displacement (w) to compute the nonlinear terms in the Green-Lagrange strains and the
initial stress matrix. They found that a linear interpolation for w is preferable with DLTP
and that the DCT element (combination of the DKT and CST elements), which gives an
over stiff solution when a cubic interpolation is used for w, could be made softer by using a
linear interpolation for w. Fafard et al. [49] also found that in the case of ULF2 the solution
failed to converge when very large initial increments were used.
The majority of the flat shell elements mentioned above use either the CST or the LST
elements to represent the membrane action. Hence those flat shell elements suffer from
in-plane rotational singularity. Though the singularity can be removed by associating a
fictitious stiffness with the in-plane rotational degree of freedom, the presence of drilling
degrees of freedom is considered essential for the analysis of shells, and elements with such
drilling degrees of freedom have been found to exhibit improved membrane response and
also a reduced sensitivity to element distortion [50].
Madenci and Barut [51] presented a flat shell element for static analysis of composite
structures, obtained by combining the flat triangular membrane and bending elements
developed by Bergan and Felippa [6] and Felippa and Bergan [36], respectively, which are
based on the free-formulation originally suggested by Bergan and Nygard [7]. Madenci and
Barut [51] employed an updated Lagrangian formulation in which the rigid body modes are
removed using the procedure given by Hsiao [45]. The presence of in-plane rotational degrees
of freedom in the formulation of Madenci and Barut [51] eliminates the in-plane rotational
singularity.
Oral and Barut [52] presented a shear-flexible flat shell element for large deflection
and instability analysis of isotropic plates and shells by combining the Allman triangular
element and the MIN3 plate bending element of Tessler and Hughes [31]. They employed an
updated Lagrangian formulation in which the rigid body modes are removed from the total
displacements using the procedure given by Hsiao [45].
In all the above mentioned flat shell elements except that of Bathe and Ho [44], Fafard
et al. [49], Meek and Tan [47] and Poulsen and Damkilde [48], large rotations are handled by
removing the rigid body modes from the total or incremental displacements. Several large
rotation formulations based on isoparametric elements and a total Lagrangian formulation
have been presented, in which the large rotations of the shell director are described by using
the large rotation matrix parametrized by 3 rotation components (see Simo and Fox [53],
Parisch [54], Ibrahimbegovic [55] for details) or by using nonlinear trigonometric functions
parametrized by 2 rotation components (see Surana [56], Brank [57] for details).
Onate et al. [58] presented a triangular flat shell element for geometrically nonlinear
analysis of isotropic shells. Their flat shell element is a combination of the LST element
and a plate bending element (based on Reissner-Mindlin theory) with linear interpolation
for the transverse displacement using corner nodes, and linear interpolation for the normal
rotations and transverse shear strains using mid-side nodes. Their formulation is applicable
for both thick and thin shell regimes. Onate et al. [58] have mentioned that their nonlinear
formulation is based on Simo’s shell theory. No details regarding the formulation as applied
to the flat shell element are presented by Onate et al. [58] nor have they presented any
examples involving large rotations to demonstrate the accuracy of the formulation.
The formulation of the flat shell element STRI3 of the commercial finite element package
ABAQUS (combination of DKT and CST elements) is also based on a description of the
motion of the shell director using the large rotation matrix (parametrized by 2 rotation
components). Not much information is available in the existing literature regarding the
formulation or the accuracy of the STRI3 element either for problems involving very large
rotations.
1.3 Objective and Scope of Present Research

The primary aim of the present research is to develop an inexpensive thin shell element
for two main applications involving thin large flexible structures : 1) control of thermal
deformations of a spherical mirror segment used in an astronomical telescope and 2) analysis
of large inflatable structures.

Flat shell elements are chosen for the analysis due to the advantages mentioned in section
1.1. A triangular element is an obvious choice since it can be used to model arbitrary
geometries. Further triangular elements with only corner connections are desired, as data
preparation becomes tedious in the case of an element with nodes at the mid-sides and
interior. The flat shell element used in the present research is a combination of the DKT
plate bending element of Batoz et al. [16] and a membrane element similar to the Allman
element [3], but derived from the LST element using the transformation given by Cook [5].
A membrane element similar to the Allman element is chosen as it does not suffer from in-
plane rotational singularity and has better membrane response than the CST element which
has been predominantly used in the flat shell formulation available in the existing literature.
The DKT element is chosen as it has been well established as an efficient and reliable thin
plate bending element.
Carpenter el al. [10] combined the one-point integrated Allman element and the DKT
plate bending element for linear analysis of isotropic shells. This underintegrated element
gave better results compared to the fully integrated element, but the reduced order
integration resulted in spurious zero energy modes. They claim that these spurious zero
energy modes are of little consequence in most practical applications and far less troublesome
than the in-plane rotational singularity. It is not well known under what circumstances the
existence of spurious mode could be troublesome. Hence such elements may not always be
reliable.
Fish and Belytschko [11] combined the stabilized, one-point integrated Allman element
and the DKT plate bending element for linear static and buckling analysis of isotropic
shells. They found that in the case of linear static analysis of the hemispherical shell the
underintegrated element gave better results than the fully integrated element even for coarse
meshes. They have shown results only up to a mesh of 128 elements which may not be
sufficient enough to accurately model a hemispherical shell. The trend of the results presented
by Fish and Belytschko [11] for linear static analysis shows that the element would over
predict the results for finer meshes or in other words the underintegrated element is too
soft. They have not fully established the convergence characteristics of the element. In the
case of buckling analysis the derivatives of the transverse displacement that are required for
computing the geometric stiffness matrix were obtained by integrating the curvature field.
In an attempt to soften the element, Fish and Belytschko [11] also employed a ont-point
integration scheme to compute the geometric stiffness matrix. Nothing can be concluded
regarding performance of such a geometric stiffness matrix in the case of nonlinear response
of general shells, as they have presented results only for linear buckling analysis.
Chen [59] combined the DKT plate bending element and the membrane element of Allman
[4] with cubic interpolation functions for the in-plane displacement fields, for linear analysis
of isotropic shells. A 6-point integration scheme is required [59] to evaluate the stiffness
matrix of the cubic membrane element of Allmanm [4]. In an attempt to improve the
performance of the element, Chen [59] used a reduced order (four-point) scheme. Chen [59]
compared the performance of the 4-point and 6-point integrated element with that of the
CST element (in combination with the DKT plate bending element), for the Scordelis-Lo
roof, pinched cylinder and the hemispherical shell problems. The 4-point integrated element
showed rapid convergence compared to that of the 6-point integrated element and the CST
element in the case of the Scordelis-Lo roof. In the case of the pinched cylinder the 4-point
integrated element did not show any improvement over the 6-point integrated element and
both gave over soft results. In the case of the hemispherical shell, the performance of the
4-point integrated element was superior to that of the 6-point integrated element, but the
convergence was not as rapid as in the case of the CST element. Better performance of the
CST element in the case of the hemispherical shell has also been noticed by Cook [15] and
Carpenter el al. [10]. The reason for such superior performance of the CST element in the
case of the hemispherical shell alone is not known.
Ertas et al. [60] presented a three-noded triangular flat shell element, termed AT/DKT,
which is a combination of the DKT plate bending element and a membrane element similar
to the Allman triangular element, but obtained from the LST element by applying the
transformation given by Cook [5]. They compared the performance of the AT/DKT element
for static analysis of anisotropic cantilever plates to that of the STRI3 element of ABAQUS.
The results presented are all related to static analysis of flat plates. The results of AT/DKT
and STRI3 presented by Ertas et al. [60] do not show excellent agreement and also have some
sign errors. They are not sufficient to ascertain that the element would converge towards
exact solutions in the case of static and dynamic response of composite shells.
In the present research, the flat shell element presented by Ertas et al. [60] is first
rigorously tested for linear static response analysis of laminated plates and shells to verify its
accuracy and determine the convergence characteristics, and is extended for free vibration,
linear thermal/piezoelectric and geometrically nonlinear analysis which include snap-back,
snap-through, thermal post buckling, dynamic buckling and the follower effects of the
pressure loads (Kapania and Mohan [61], Mohan and Kapania [62], Mohan and Kapania
[63]).
The major drawback of the DKT element is that the transverse displacement field is not
explicitly defined in the interior of the element. Hence a consistent mass matrix and the
derivatives of the transverse displacement that are required to obtain the geometric stiffness
matrix (initial stress matrix) in the case of buckling and postbuckling analysis cannot be
readily obtained. Free vibration analysis could be performed using a lumped mass matrix,
but the analysis is expensive, especially if a large number of modes is required to be extracted,
due to poor rate of convergence. In the present research, free vibration analysis is performed
using both a lumped mass matrix and a so called consistent mass matrix, in order to compare
the two methods. The consistent mass matrix is obtained using the nine term polynomial
in area coordinates given by Specht [37] to represent the transverse displacement and the
standard quadratic basis in area coordinates for the in-plane displacements. Unlike the
lumped mass matrix, the rotary inertias are included in the consistent mass matrix.
In the case of thermal analysis the temperature variation along the thickness of the
element and across the plane of the element is assumed to vary linearly in terms of the
nodal temperatures. The formulation can be applied for the analysis of structures excited
by an electric field by replacing the coefficients of thermal expansion by the electric field
constants and the thermal field by the electric field, assuming a linear relation exists between
the induced strain and the applied electric field in terms of the electric field constants.
A structure with any number of actuators bonded or embedded in a substrate can be
analyzed provided the overall thickness of the structure is small enough to neglect the shear
deformations. The electro-thermo-mechanical coupling has not been included in formulation,
but can be easily incorporated as most of the strain-displacement matrices required are
already available.
The updated Lagrangian formulation employed in the present study is based on Green-
Lagrange strain measures and PK2 (Second Piola-Kirchhoff) stress measures. Since the
transverse displacement (w) is not explicitly defined over the interior of the DKT element, the
determination of the derivatives of the transverse displacement (w,x, w,y ) which are required
to compute the initial stress matrix is not straightforward. The derivatives (w,x, w,y ) are
determined using a linear interpolation field for w in terms of the nodal values. This kind of
simpler interpolation for the transverse displacement has been widely used in many updated
Lagrangian formulations in the existing literature that employ a flat shell element. The
use of such linear interpolation for the displacement field to compute the geometric stiffness
matrix is justified probably because most of the deformation of the element is made of rigid
body rotations.
The element is first applied to study the control of thermal deformations of a spherical
mirror segment (Kapania et al. [64]). This kind of study was required for the design of
a next generation space telescope proposed by Jakubowski et al. [65], which employs a
multi-segmented primary mirror. The feasibility of controlling the surface distortions of the
mirror due to arbitrary thermal fields, using piezoelectric strips and force control actuators is
studied using a finite element model of the mirror made up of the flat shell elements. In the
case of the piezoelectric strips, the induced strain due to an externally applied voltage across
the thickness of the strip bonded to the rear surface of the mirror, is used to control the
surface distortions of the mirror. In the case of the force actuators the force exerted by the
actuators normal to the rear surface of the mirror is used to control the surface distortions.
The optimal voltages and the forces to be applied to the strips and actuators, respectively,
are determined by minimizing the rms surface figure error of the mirror segment, which is
a measure of the deviation of the deformed mirror surface from the desired surface. Since
the piezoelectric strips have very light weight, a large number of such strips could be used
without increasing the overall weight or stiffness of the mirror, unlike the force actuators
which require a heavy backup structure which is not desirable for space applications. Since
the piezoelectric strips are bonded to the rear surface of the mirror, they cannot be relocated.
Hence a configuration of strips with predetermined locations will have to be used to control
arbitrary surface distortions. A study is also conducted in order to compare the performance
of a set of evenly placed strips and (near) optimally placed strips. The optimal locations of
the piezoelectric strips are determined using a heuristic technique. Since the force actuators
are used for both support and control, changing the locations of the actuators will result in
different support configurations and hence different thermal deformations. In other words
the problem does not remain the same by changing the locations of the actuators. Hence no
attempt is made to determine the optimal locations.
The second application for which the element is intended to be used is the analysis of
large inflatable structures which are expected to undergo large deformations. The updated
Lagrangian formulation (Mohan and Kapania [62]) has been primarily developed to analyze
large thin inflatable structures. When analyzing structures subject to pressure loads, the
change in the direction of the pressure loads (follower action) should be taken into account in
order to accurately compute the pressure load vector. As a result of the follower action, the
pressure load vector is a function of the displacements and hence contributes to the tangent
stiffness matrix as is the case with the thermal load. In the case of the thermal load, this
contribution, is similar to the initial stress matrix and hence no additional effort is required
to compute this contribution. In the case of the pressure load, this contribution called the
pressure stiffness is in general unsymmetric and its determination is not straightforward but
can be systematically derived from the principle of virtual work.
The literature on large deformation analysis, using the finite element method, of general
shells under displacement-dependent pressure loads is very scarce. Hibbitt [66] presented
a brief derivation of the load stiffness associated with some follower forces like pressure,
centrifugal force, frictional drag and cable force using the principle of virtual work. Hibbitt
showed that the pressure stiffness matrix can be split into two parts, a symmetric part
corresponding to the interior and an antisymmetric part corresponding to the boundary of
the structure, respectively. Hibbitt also showed that in the case of uniform pressure, with
displacements prescribed on the boundaries, the pressure stiffness matrix becomes symmetric
as the boundary terms vanish. When there are free boundaries, Hibbitt suggested that a
symmetric pressure stiffness matrix defined as the average of the pressure stiffness matrix
and its transpose could be used provided the elements near the free boundary are made
very small. Neither details of the formulation nor any numerical examples are presented by
Hibbitt.
Loganathan et al. [67], using a curved triangular shell element, presented linear buckling
analyses of a thin circular arch and a cylindrical shell under uniform external pressure. They
demonstrated that the inclusion of the pressure stiffness gave accurate values of the critical
pressure.
Schweizerhof and Ramm [68] presented a total Lagrangian formulation applicable for
isoparametric beam and shell elements, including the follower effects of the displacement
dependent pressure loads. They considered two types of pressure loads: 1) space attached
load for which the distribution of the load magnitude is a function of the coordinates in
the deformed configuration; and 2) body attached load for which it is a function of the
coordinates in the initial undeformed configuration. In both cases the pressure load acts
normal to the deformed configuration. They showed that, for both types, the boundary
terms of the pressure stiffness matrix are in general skew-symmetric, and the domain
terms are always symmetric in the case of space attached loads, are symmetric for body
attached uniform loads and are skew-symmetric for body attached nonuniform loads. They
also outlined different types of boundary conditions under which the boundary terms are
symmetric for both types of loads and suggested that the most efficient way to symmetrize
the pressure stiffness matrix is to simply neglect the skew-symmetric terms. They came to
interesting conclusions that flutter instability can only be detected using the unsymmetric
pressure stiffness matrix, whereas in the case of static stability analysis the unsymmetric
part has a minor influence on the critical load. They also remarked that the effect of
totally neglecting the load/pressure stiffness matrix in determining the nonlinear response
in some cases is only an increase in the number of equilibrium iterations. This statement is
incorrect. From the author’s experience, if the pressure stiffness matrix is totally neglected,
the equilibrium iterations may not converge at all, especially in the regime of large rotations.
Yuan and Liang [69] presented the derivation of the pressure stiffness for a three-node
axisymmetric shell element. Iwata et al. [70] presented a general form of the pressure stiffness
matrix in a Cartesian coordinate system by relating, using the relationship given by Malvern
[71], the area vector in the deformed configuration to that in the undeformed configuration.
The derivation presented by Iwata et al. [70] does not distinguish clearly between the area
elements in the deformed and undeformed configurations. They only presented results of a
linear buckling analysis.
Gruttmann and Taylor [72] presented a membrane formulation including the pressure
stiffness matrix. Their derivation of the pressure stiffness matrix is also based on relating
the area vector in the deformed and undeformed configurations. Their pressure stiffness
matrix turns out to be skew symmetric. They presented several numerical examples on the
nonlinear response of rubber like membrane shells.
Christensen [73] studied the effects of pressure on the dynamics of structures stiffened
due to high internal pressure such as the Space Shuttle solid rocket booster and the solid
rocket motor. He presented a general formulation of the pressure stiffness for triangular and
quadrilateral flat shell elements based on the assumption that the element deforms as a rigid
body and hence used a linearly varying field for the translations. All the terms involving the
nodal translational degrees of freedom in the expression for the pressure stiffness are neglected
assuming that the nodal translations are much smaller than the element dimensions. This
assumption may not be true in the case of problems involving large deformations. Christensen
found that the natural frequencies of the rocket booster and motor, with the inclusion of the
pressure stiffness, were in good agreement with those of the experimental results.
In the present research, large deformation analysis of general shells under deformation
dependent pressure loads is performed using the updated Lagrangian formulation of the
flat shell element (Mohan and Kapania [62]), by including the follower effects (Mohan and
Kapania [63]). The pressure stiffness matrix and the deformation dependent pressure load
vector are derived in a Cartesian coordinate system using the principle of virtual work. As
in the case of the formulation by Christensen [73] and Schweizerhof and Ramm [68], the
contribution of the rotational degrees of freedom to the pressure stiffness matrix and the
pressure load vector is neglected. Unlike the formulation presented by Christensen [73], the
displacement dependent terms in the pressure stiffness matrix and also in the pressure load
vector are not neglected. Since the major part of the deformation of the element in the
case of large deformation analysis is made up of rigid body motions, a linear interpolation
field is assumed for the translational degrees of freedom as was done by Christensen [73]. A
symmetric pressure stiffness matrix as proposed by Hibbitt [66] is employed in the present
study.
The details of the finite element formulation are presented in Chapter 2. Several numerical
examples on static, free vibration, thermal and geometrically non-linear analysis of laminated
plates and shells are presented in Chapter 3 to demonstrate the validity of the present
formulation. The first application of the element, namely the control of thermal deformation
of a spherical mirror segment, is presented in Chapter 4.
The element is currently being used to analyze a large inflatable structure which is being
considered by the US Army to be used as a transportable tent to accommodate Army vehicles
and personnel. The tent is made up of arches stiffened by inflation and membranes that are
connected to the arches and held under tension. A model of the tent structure made of two
arches and a membrane is shown in (Fig. 1.1). The deformation of the tent under wind and
snow loads is currently being studied. The results of the preliminary analysis are given in
Figure 1.1: Tent structure
Chapter 5.
Chapter 2
Finite Element Formulation
The flat shell element used in this study is obtained by combining the Discrete Kirchhoff
Theory (DKT) plate bending element of Batoz et al. [16] and a membrane element similar
to the Allman triangle [3], but derived from the well known Linear Strain Triangular (LST)
element. The derivation of the stiffness matrices in explicit form is not presented here. An
elegant explicit formulation of the LST element is given by Subramanian and Bose [74].
An alternate explicit formulation of the DKT element along with a FORTRAN code are
given by Jeyachandrabose [75]. A brief description of the LST and DKT elements is given
first. Then the flat shell formulation and the solution techniques used are presented. Linear
and non-linear formulations are dealt together. The updated Lagrangian formulation for a
generally laminated structure is presented in detail, from which the governing equations for
the linear analysis can be easily obtained. Only the solution techniques used to solve the
system of non-linear equations are presented in detail since such techniques, though very
widely used, have not been very well documented in the existing literature, especially in the
case of thermal and pressure loads. Solution techniques like the subspace iteration and the
derivation of the consistent mass matrix are not reproduced as they are well documented in
many standard text books.
2.1 Membrane Formulation

The membrane element used in the present study has three nodes and three degrees of
freedom at each node (two inplane displacements and an inplane rotation). The degrees of
24
CHAPTER 2. FINITE ELEMENT FORMULATION 25
freedom of the membrane element are identical to those of the Allman triangular element,
but the strain-displacement relations are derived from those of the LST element (which
has three corner and three midside nodes and two inplane displacements at each node) by
eliminating the midside displacements in terms of the corner displacements and rotations
using the transformation suggested by Cook [5].
In the case of the LST element, the in-plane displacement fields u and v along the local x
and y axes, respectively, are represented by the standard quadratic basis in area coordinates
as follows
6
6

u= Pi ui , v= Pi vi (2.1)
i=1 i=1
where ui and vi are the nodal values of the in-plane displacements and Pi are the standard
quadratic shape functions in area coordinates ξ and η given by
⎧ ⎫
⎪
⎪
2(1 − ξ − η)(0.5 − ξ − η) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ξ(2ξ − 1) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎨ ⎪
⎬
η(2η − 1)
{P } = ⎪ ⎪
(2.2)
⎪
⎪
⎪
4ξ(1 − ξ − η) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ 4ξη ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎩ ⎪
⎭
4η(1 − ξη)
The local x-axis is taken along the side 1-2 of the element with the origin at node 1. The
local y-axis is perpendicular to the local x-axis and lies in the plane passing through the
three nodes. The shape functions in {P } are arranged in such an order that the nodes 4, 5
and 6 are at the mid point of the sides 1-2, 2-3 and 3-1, respectively.
The normal and tangential displacements un and ut along a typical element side 1-2 for
example, as given by Allman [3] are,
s s s s
un (s) = (1 − )un1 + un2 + (1 − )(ω2 − ω1 ) (2.3)
L12 L12 2 L12
s s
ut = (1 − )ut1 + ut2 (2.4)
L12 L12
where s is the coordinate along edge 1-2, L12 is the length of the edge 1-2, un1 , ut1, un2 , ut2
are the normal and tangential edge displacements at the nodes 1 and 2, respectively, ω1 and
ω2 are the drilling degrees of freedom at nodes 1 and 2, respectively. The drilling degree of
freedom ω is not exactly the true inplane rotation, but can be treated as the variation of
the edge normal displacement un with respect to the edge coordinate s or in other words
ω = ∂un/∂s. The true inplane rotation about the normal to the plane of the element is given
by
1
Ω = (∂v/∂x − ∂u/∂y) (2.5)
2
Using straightforward algebra, it can be shown that Ω and ω are related by
3
Ωi − Ωo = (ωi − ωo ) (2.6)
4
where Ωi are the true rotations at the vertices, Ω0 is the average value of the true nodal
rotations, and ωi and ω0 are the corresponding values of the drilling degree of freedom. The
edge normal and tangential displacements un and ut can be expressed as
un = u cos(γ12 ) + v sin(γ12) (2.7)
ut = −u sin(γ12) + v cos(γ12 ) (2.8)
where γ12 is the angle between the outward normal to the edge 1-2 and the positive direction
of the local x-axis given by
cos(γ12 ) = (y2 − y1)/L12 , sin(γ12) = (x1 − x2)/L12 (2.9)
The midside displacements u4 and v4 can now be expressed in terms of the corner
displacements and rotations as
1 1
u4 = u1 + u2 + (y1 − y2 )(ω1 − ω2 )/8 (2.10)
2 2
1 1
v4 = v1 + v2 + (x2 − x1)(ω1 − ω2 )/8 (2.11)
2 2
Similar expressions can be obtained for the midside displacements of the other two edges.
In the updated Lagrangian formulation discussed later in this section, the inplane strain-
displacement relations are first obtained in terms of the degrees of freedom of the LST
element, using the shape functions of the LST element and are expressed in terms of
the degrees of the Allman element by eliminating the mid-side displacements using the
expressions shown above. The shape functions proposed by Allman [5] are not used in any
computations. The in-plane strain-displacement relations are presented later in this Chapter.
Z Z
Deformed normal Deformed normal
y x
o Y X o
Figure 2.1: Positive directions of βx and βy
2.2 Plate Bending Formulation

The formulation of the DKT plate bending element proceeds first by assuming that the
transverse shear deformations are present. In other words the normal to the undeformed
middle surface of the element remains straight after deformation, but is not necessarily
normal to the deformed middle surface. With this assumption and considering that the
rotations of the normal are small, the displacement components u, v, w of any point x, y, z
in a local coordinate system can be expressed as
u = zβx (x, y); v = zβy (x, y); w = w(x, y) (2.12)
where βx and βy are the rotations of the normal to the undeformed middle surface in the x-z
and y-z planes, respectively. The positive senses of βx and βy are shown in Fig. 2.1. The
bending strains (varying linearly through the thickness) can be expressed as
{ b } = z{κ} (2.13)
where {κ} is the vector of curvatures given by

⎧ ⎫
⎪
⎪ βx,x ⎪
⎪
⎨ ⎬
{κ} = ⎪ βy,y ⎪
(2.14)
⎪
⎩ ⎪
⎭
βx,y + βy,x
The transverse shear energy is neglected and the strain energy consisting only of the bending
energy is expressed as
1
U= {κ}T [D]{κ} dA (2.15)
2 A
where A is the area of the undeformed middle surface of the element and [D] is the standard
matrix of elastic constants. Since the strain energy involves only the first derivatives of the
normal rotations, the shape functions used to represent the normal rotations need to satisfy
only Co continuity. Hence the normal rotations can be represented using the standard
quadratic shape functions in area coordinates as
6

βx = βxi Pi (2.16)
i=1
6

βy = βyi Pi (2.17)
i=1
where βxi and βyi are the values of the normal rotations at the corners and mid point of
element edges. The nodal degrees of freedom of the Kirchhoff plate elements in general are
the transverse displacement w and the slopes θx = w,y and θy = −w,x . In order to derive
the discrete equations of equilibrium in terms of the unknown nodal degrees of freedom of
the element, the unknowns βxi and βyi should be expressed in terms of the nodal degrees of
freedom wi , θxi and θyi . The 12 equations that relate the normal rotations at the corners
and mid-sides to the degrees of freedom of the element are obtained by applying the discrete
Kirchhoff assumptions which are explained in detail by Batoz et al. [16]. After applying the
discrete Kirchhoff constraints, the normal rotations βx and βy can be expressed as
1 1
βx = [Hx (ξ, η)]{adkt} βy = [Hy (ξ, η)]{adkt } (2.18)
2A 2A
where {adkt } is the vector of nodal degrees of freedom of the DKT element given by {adkt}T =
{w1 θx1 θy1 w2 θx2 θy2 w3 θx3 θy3 }, and [Hx ] and [Hy ] are the vectors of shape functions given
by Batoz et al. [16]. These shape functions have been reordered such that the nodes 4, 5
and 6 are at the midpoints of the sides 1-2, 2-3 and 3-1, respectively. This reordering has
been done in order to match the order of the nodes with that of the LST element to avoid
confusion while coding. The shape functions [Hx ] and [Hy ] after reordering are given by
⎧ ⎫ ⎧ ⎫
⎪
⎪ 1.5(a4P4 − a6P6 ) ⎪ ⎪ ⎪
⎪ 1.5(d4 P4 − d6 P6 ) ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪ ⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪
b 4 P 4 + b 6 P 6 ⎪
⎪
⎪
⎪
⎪
⎪
−P1 + e4 P4 + e6 P6 ⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ P − c P − c P ⎪
⎪ ⎪
⎪ −b4P4 − b6P6 ⎪
⎪
⎪
⎪ 1 4 4 6 6 ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ 1.5(a 5 P 5 − a 4 P 4 ) ⎪
⎪ ⎪
⎪ 1.5(d5 P5 − d4 P4 ) ⎪
⎪
⎨ ⎬ ⎨ ⎬
Hx = ⎪ b P + b P
5 5 4 4 Hy = ⎪ −P1 + e5 P5 + e4 P4 ⎪
⎪ ⎪
⎪ ⎪ ⎪
⎪ ⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪
⎪
P 1 − c 5 P5 − c 4 P4 ⎪
⎪
⎪
⎪
⎪
⎪
−b 5 P5 − b 4 P4 ⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪
⎪ − ⎪ ⎪ 1.5(d6 P6 − d5 P5 ) ⎪
⎪
⎪ 1.5(a P
6 6 a 5 5 ⎪
P ) ⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎪ ⎪ ⎪
⎪
⎪
⎪ b 6 P6 + b 5 P 5
⎪
⎪
⎪
⎪
⎪
⎪ −P1 + e6 P6 + e5 P5 ⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪ ⎪
⎪ ⎪
⎪
⎩ ⎭ ⎩ ⎭
P1 − c6P6 − c4 P4 −b6P6 − b5P5
where
2
ak = −xij /lij
3
bk = x y /l2
4 ij ij ij
ck = ( 14 x2ij − 12 yij2 )/lij
2
2
dk = −yij /lij
ek = ( 14 yij2 − 12 x2ij )/lij
2
2
lij = x2ij + yij2
xij = xj − xi
yij = yj − yi
where k = 4, 5, 6 corresponding to the sides ij = 1-2, 2-3, 3-1, respectively, and xi , yi are the
nodal coordinates in the local coordinate system. As mentioned before, the derivation of the
stiffness matrix in explicit form is not presented here. Only the shape functions of the LST
and DKT elements are required to derive the governing non-linear equations of equilibrium
of a generally laminated shell structure, which is presented in the next section.
2.3 Updated Lagrangian Formulation
2.3.1 Static analysis

The nonlinear equations of equilibrium governing the static response of a generally laminated
structure are derived by applying the Principle of Virtual Work for a typical element using a
local coordinate system as mentioned in section 2.1. It is assumed that a known equilibrium
configuration C1 at some time t = t1 is available and that a new configuration C2 at some

time t = t2 has to be determined. This is achieved by solving the nonlinear equations
of equilibrium using the Newton-Raphson method, which is a standard iterative solution
technique. Both C1 and C2 denote a local co-ordinate system. The Principle of Virtual
Work can be expressed for a single element as follows:
δWe = δWi (2.19)
where δWe is the virtual work done by external forces and δWi is the virtual work done by
internal forces. The internal virtual work is given by

δWi = δ{ }T {s} dV (2.20)
C1
where V is the volume of the element in C1 , { } is the vector of incremental Green-Lagrange

strains and {s} is the vector of Second Piola-Kirchhoff (PK2) stresses. The PK2 stresses at
any time t1 < t < t2 during the solution process can be decomposed as
{s} = {σ} + {Δs} (2.21)
where {σ} is the vector of Cauchy stresses in C1 at time t = t1 and {Δs} is the vector of
incremental PK2 stresses. The basic difference between the Cauchy stresses and the PK2
stresses is that the Cauchy stresses are measured in the current configuration, whereas the
PK2 stresses are measured in the reference configuration. The term “current configuration”
is used to denote any configuration between C1 and C2 obtained during the solution process.
The incremental Green-Lagrange strains and PK2 stresses in any layer k of a laminated
structure are related by the constitutive equations given by
{Δs} = [Q̄k ]{ } (2.22)
where [Q̄k ] is the standard matrix of elastic constants transformed to the local element
coordinate system and is given by
⎡ ⎤⎡ ⎤⎡ ⎤T
c2 s2 −2cs q11 q12 0 c2 s2 −2cs
⎢ ⎥⎢ ⎥⎢ 2 ⎥
[Q̄]k = ⎢
⎣s
2
c2 2cs ⎥ ⎢
⎦ ⎣ q12 q22 0 ⎥ ⎢
⎦⎣s c2 2cs ⎥
⎦ (2.23)
cs −cs c2 − s2 0 0 q66 cs −cs c2 − s2
where q11 = E1 /(1 − ν12ν21 ), q12 = ν12E2 /(1 − ν12 ν21), q22 = E2 /(1 − ν12 ν21), q66 = G12 ,
c = cos(θ), s = sin(θ) and E1 , E2 , ν12, G12 are the properties in the principal material
directions of the k th layer and θ is the angle which the E1 direction of the k th layer makes
with the local x-axis, measured counter clock wise from the local x-axis.
At the end of the current step (t = t2 ) the PK2 stresses in C1 , computed using Eqs. 2.21
and 2.22, are converted to Cauchy stresses in C2 by a series of transformations involving
the deformation gradient tensor and the rotation tensor (Kleiber [76]), but for small strains
it can be shown (Farfard et al. [49] ) that at the end of the current step the PK2 stresses
computed in C1 are equal to the Cauchy stresses in C2 .
Assuming that the thickness of the structure and the incremental rotations are moderate,
the variation of strain through the thickness can be expressed as
{ } = {e} + z {κ} −{ 0} (2.24)
where {e}, {κ} and { 0} are the vectors of incremental membrane strains, bending strains
and thermal strains, respectively, and are given by
⎧ ⎫
⎪
⎪ u,x + 1/2(u2,x + v,x
2 2
+ w,x ) ⎪
⎪
⎨ ⎬
{e} = ⎪ v,y + 1/2(u2,y + v,y
2 2
+ w,y ) (2.25)
⎪ ⎪
⎪
⎩ ⎭
u,y + v,x + u,x u,y + v,x v,y + w,xw,y
{κ} = {βx,x βy,y βx,y + βy,x }T (2.26)
{ 0 } = {αx αy αxy }Tk ΔT (2.27)
where u, v, w are the incremental displacements of the mid-plane of the element, βx and βy
are the incremental rotations of the normal to the undeformed mid-plane in the local x − z
and y − z planes, respectively, ΔT is the prescribed temperature increment, and αx , αy and
αxy are the coefficients of thermal expansion with respect to the element local coordinate
system for the k th layer given by
⎧ ⎫ ⎡ ⎤ ⎧ ⎫
⎪
⎪ αx ⎪⎪ c2 s2 −cs ⎪
⎪ α1 ⎪
⎪
⎨ ⎬ ⎢ 2 ⎥ ⎨ ⎬
α = ⎢ s c2
cs ⎥ α (2.28)
⎪ y
⎪ ⎣ ⎦ ⎪ 2 ⎪
⎪
⎩ ⎪
⎭ ⎪
⎩ ⎪
⎭
2 2
αxy k 2cs −2cs c − s k α12 k
where α1 , α2 and α12 are the coefficients of thermal expansion with respect to the principal
material directions of the kth layer. The internal virtual work can now be expressed as

δWi = δ{ } {s} dV =
T
δ{e} {N} + δ{κ} {M} dA
T T
(2.29)
C1 C1
where A is the area of the mid-plane of the element in C1, and {N} and {M} are the force
and moment resultants, respectively:
h
2
{N, M} = {s}(1, z) dz (2.30)
− h2
where h is the thickness of the laminate. Using Eqs. 2.21 and 2.22, the force and moment
resultants can be decomposed as
{N} = {N1 } + {ΔN m } − {ΔN 0 } (2.31)
{M} = {M1 } + {ΔM m } −{ ΔM 0} (2.32)
where {N1 }, {M1 } are the force and moment resultants in C1 at time t = t1, {ΔN m },
{ΔM m } are the incremental force and moment resultants due to mechanical stresses and
{ΔN 0 }, {ΔM 0 } are the incremental force and moment resultants due to thermal stresses:
h
2
{N1, M1 } = {σ}(1, z) dz (2.33)
−h
2
h
2
{ΔN m } = [Q̄k ] ({e} + z{κ}) dz = [A]{e} + [B]{κ} (2.34)
−h
2
h
2
{ΔM } =
m
[Q̄k ] ({e} + z{κ})z dz = [B]{e} + [D]{κ} (2.35)
− h2
h
0 0 2
{ΔN , ΔM } = [Q̄k ] { 0} (1, z) dz (2.36)
−h
2
where [A], [B], [D] are the standard matrices of elastic constants. In the present study all
the area integrals are evaluated using a 3 point integration scheme in area coordinates (Yang
[2]). The integral over the thickness is computed by summing the contribution from each
layer of the laminate. The first variation of the membrane strains can be expressed as
⎧ ⎫
⎪
⎪ δu,x + u,xδu,x + v,xδv,x + w,x δw,x ⎪
⎪
⎨ ⎬
{δe} = ⎪ δv,y + u,y δu,y + v,y δv,y + w,y δw,y (2.37)
⎪ ⎪
⎪
⎩ ⎭
δu,y + δv,x + u,x δu,y + u,y δu,x + v,xδv,y + v,y δv,x + w,xδw,y + w,y δw,x
The above equation can be rearranged as
{δe} = [G1 ]{δu,x δu,y δv,x δv,y δw,x δw,y }T (2.38)

where ⎡ ⎤
1 + u,x 0 v,x 0 w,x 0
⎢ ⎥
[G1 ] = ⎢
⎣ 0 u,y 0 1 + v,y 0 w,y ⎥
⎦ (2.39)
u,y 1 + u,x 1 + v,y v,x w,y w,x
The in-plane displacements u and v can be expressed in terms of the nodal quantities as

u {P }T 0
= {alst} (2.40)
v 0 {P }T
where {alst } is the vector of nodal degrees of freedom of the LST element given by
{alst } = {u1, u2 , u3, u4, u5 , u6, v1, v2, v3 , v4, v5, v6} and {P } is the vector of quadratic shape
functions in area co-ordinates given by Eq. 2.2. The derivatives of the in-plane displacements
are given by ⎧ ⎫ ⎡ ⎤
⎪
⎪ u,x ⎪
⎪ {P,x }T 0
⎪
⎪ ⎪
⎪ ⎢ ⎥
⎪
⎨ u,y ⎪
⎬ ⎢ {P,y }T ⎥
⎢ 0 ⎥
= ⎢ ⎥ [Tlst]{a} = [Blst ]{a} (2.41)
⎪
⎪ v,x ⎪
⎪ ⎢ 0 {P,x }T ⎥
⎪
⎪ ⎪
⎪ ⎣ ⎦
⎪
⎩ ⎪
⎭
v,y 0 {P,y }T
where {P,x } and {P,y } are the vectors of derivatives of the shape functions with respect to
the local x and y co-ordinates, respectively, and are given by

1
{P,x } = b2 {P,ξ } + b3{P,η } (2.42)
2A

1
{P,y } = c2 {P,ξ } + c3 {P,η } (2.43)
2A
where 2A = b2c3 − c2 b3, b2 = y3 − y1, b3 = y1 − y2, c2 = x1 − x3, c3 = x2 − x1 . The matrix
[Tlst] is used to express the nodal degrees of the LST element {alst } in terms of the nodal
degrees of freedom of the shell element {a} as {alst } = [Tlst]{a}, where {a}T = {u1 , v1, w1,
θx1 , θy1 , θz1 , u2, v2, w2 , θx2 , θy2 , θz2 , u3, v3 , w3, θx3 , θy3 , θz3 }. The transformation matrix [Tlst] is
given by
⎡ ⎤
1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
⎢ ⎥
⎢ 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0.5 0 0 0 0 b3/8 0.5 0 0 0 0 −b3/8 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 0.5 0 0 0 0 0.5 0 0 0 0 −b1/8 ⎥
⎢ b1 /8 ⎥
⎢ ⎥
⎢
⎢ 0.5 0 0 0 0 −b2/8 0 0 0 0 0 0 0.5 0 0 0 0 b2/8 ⎥⎥
[Tlst] = ⎢ ⎥
⎢ ⎥
⎢ 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0.5 0 0 0 c3 /8 0 0.5 0 0 0 −c3/8 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 0 0 0 0 0 0.5 0 0 0 c1 /8 0 0.5 0 0 0 −c1 /8 ⎥
⎣ ⎦
0 0.5 0 0 0 −c2/8 0 0 0 0 0 0 0 0.5 0 0 0 c2/8
In the case of the DKT element the transverse displacement w is not defined explicitly
over the interior of the element. The derivatives w,x and w,y at the integration points
can be obtained using a linear interpolation in terms of the nodal values, but this kind
of interpolation required a very fine mesh and small load steps, probably due to the non-
conformal nature of the derivatives along the element sides. Past experience (Farfard et al.
[49]) has shown that evaluating the derivatives from a linear interpolation for w using the
nodal values gives better results. In the present study the same approach is followed. In
area coordinates, w is expressed as w(ξ, η) = (1 − ξ − η)w1 + ξw2 + ηw3 from which the
derivatives can be obtained as

w,x 1 −b2 − b3 b2 b3
= [Tw ]{a} = [Bw ]{a} (2.44)
w,y 2A −c2 − c3 c2 c3
where [Tw ] is used to express {w1 w2 w3 }T in terms of {a} as {w1 w2 w3 }T = [Tw ]{a}.
The transformation matrix [Tw ] is made of zeroes and ones, the determination of which is
straightforward. Combining Eqs. 2.41 and 2.44, the first variation of the membrane strains
can be expressed as
{δe} = [G1 ][G2]{δa} (2.45)
where [G2 ] is a matrix of size 6x18 with the first 4 rows made of [Blst ] and the last 2 rows
made of [Bw ]. The first variation of the bending strains can now be expressed as
⎡ T T ⎤
b2 Hx,ξ + b3Hx,η
1 ⎢
⎢
⎥
⎥ [Tdkt]{δa}
{δκ} = T
c2 Hy,ξ T
+ c3Hy,η = [Bdkt ]{δa} (2.46)
2A ⎣ ⎦
T T T T
c2Hx,ξ + c3 Hx,η + b2 Hy,ξ + b3Hy,η
where Hx,ξ , Hx,η , Hy,ξ , Hy,η are the derivatives of the shape functions with respect to the area
coordinates given by
⎧ ⎫
⎪
⎪ p4(1 − 2ξ) + (p6 − p4)η) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
q4(1 − 2ξ) − (q6 + q4)η) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −4 + 6(ξ + η) + r4(1 − 2ξ) − η(r6 + r4) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −p4(1 − 2ξ) + η(p5 + p4) ⎪
⎪
⎨ ⎬
Hx,ξ = q4(1 − 2ξ) − η(q4 − q5)
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
−2 + 6ξ) + r4(1 − 2ξ)) + η(r5 − r4) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −η(p6 + p5) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ η(q5 − q6) ⎪
⎪
⎪
⎪ ⎪
⎪
⎩ ⎭
−η(r6 − r5)
⎧ ⎫
⎪
⎪ −p6(1 − 2η) − (p4 − p6)ξ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
q6(1 − 2η) − (q6 + q4)ξ ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −4 + 6(ξ + η) + r6(1 − 2η) − ξ(r6 + r4) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ξ(p5 + p4) ⎪
⎪
⎨ ⎬
Hx,η = ⎪ ξ(q5 − q4)
⎪ ⎪
⎪
⎪ ⎪ ⎪
⎪
⎪
⎪
⎪
−ξ(r4 − r5) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ p6(1 − 2η) − ξ(p5 + p6) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ q6(1 − 2η) + ξ(q5 − q6) ⎪
⎪
⎪
⎪ ⎪
⎪
⎩ ⎭
−2 + 6η + r6(1 − 2η) + ξ(r5 − r6)
⎧ ⎫
⎪
⎪ t4(1 − 2ξ) + η(t6 − t4) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
1 + r4(1 − 2ξ) − η(r6 + r4) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −q4(1 − 2ξ) + η(q6 + q4) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −t4(1 − 2ξ) + η(t5 + t4) ⎪
⎪
⎨ ⎬
Hy,ξ = −1 + r4(1 − 2ξ) + η(r5 − r4)
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
−q4(1 − 2ξ) − η(q5 − q4) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −η(t5 + t6) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ η(r5 − r6) ⎪
⎪
⎪
⎪ ⎪
⎪
⎩ ⎭
−η(q5 − q6)
⎧ ⎫
⎪
⎪ −t6(1 − 2η) − ξ(t4 − t6) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
1 + r6(1 − 2η) − ξ(r6 + r4) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ −q6(1 − 2η) + ξ(q6 + q4) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ ξ(t5 + t4) ⎪
⎪
⎨ ⎬
Hy,η = ξ(r5 − r4)
⎪
⎪ ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
−ξ(q5 − q4) ⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪ − 2η) − + ⎪
⎪
⎪
⎪ t6(1 ξ(t5 t6) ⎪
⎪
⎪
⎪ ⎪
⎪
⎪
⎪
⎪
⎪ −1 + r6(1 − 2η) + ξ(r5 − r6) ⎪
⎪
⎪
⎪
⎪
⎪ ⎪
⎪
⎩ ⎭
−q6(1 − 2η) − ξ(q5 − q6)
The matrix [Tdkt ] which is used to expresses {adkt } in terms of {a} as {adkt } = [Tdkt ]{a}, is
made up of zeroes and ones, the determination of which is straightforward.
The external virtual work can be expressed as
δWe = {δa}T {f} (2.47)
where {f} is the element external force vector. The equations of equilibrium at the element
level can be expressed as
{g} = {q} − {f} = 0 (2.48)
where {q} is the element internal force vector given by

{q} = [G2 ] [G1 ] {N} + [Bdkt ] {M} dA
T T T
(2.49)
C1
The element internal and external force vectors thus obtained in the element local
co-ordinates are converted to the global co-ordinates using the standard coordinate
transformation (Zienkiewicz [13]) and are assembled to obtain the global internal force
vector q and external force vector f. In the rest of this chapter vectors and matrices in
the global coordinate system are denoted in bold face. The non-linear governing equations
of equilibrium of the entire finite element model in the global Cartesian co-ordinate system
at any time t can be written as gt = qt − ft = 0, where g is the residue or the imbalance
between the internal and external forces.
The non-linear equations of equilibrium are solved using the Newton-Raphson method,
an iterative or step-by-step process in which a linearized form of the equilibrium equations
is solved in each step. Assuming that a known equilibrium configuration exists at some time
t1, the equations of equilibrium at some time t1 + Δt can be linearized using the truncated
Taylor series expansion about the known configuration at time t1 as

∂q
gt1 +Δt = qt1 + δa − ft1 +Δt = qt1 + [K] δa − ft1 +Δt = 0 (2.50)
∂a t1
where [K] is the global tangent stiffness matrix which is obtained by transforming the local
element stiffness matrices [k] (obtained by taking the derivative of the element internal force
vector {q} in Eq. 2.49 with respect to {a}) and assembling them in the standard way. The
external force vector f could also be a function of the displacements. This case is dealt with
later in this Chapter. The derivative of the terms in Eq. 2.49 are obtained using the chain
rule δ( ) = (∂( )/∂{a})δ{a}. Using Eqs. 2.31, 2.32, 2.34, 2.35, 2.37 and 2.46, the derivatives
of {N} and {M} with respect to {a} can be readily obtained as
∂{N} ∂{M}
= [A][G1][G2] + [B][Bdkt] = [B][G1][G2] + [D][Bdkt ] (2.51)
∂{a} ∂{a}
The variation of the product [G2 ]T [G1 ]T {N} is given by
δ([G2 ]T [G1 ]T {N}) = [G2 ]T (δ[G1]T ){N} + [G2 ]T [G1]T δ{N} (2.52)
Using the special property of the matrix [G1 ] (Zienkiewicz and Taylor [77], p. 294), the
product (δ[G1]T ){N} can be expressed as
⎡ ⎤
N̂ 0 0
⎢ ⎥
(δ[G1 ]T ){N} = ⎢
⎣ 0 N̂ 0⎥
⎦ [G2 ]δ{a} (2.53)
0 0 N̂

nx nxy
where nx , ny and nxy are the components of {N} and N̂ =
nxy ny
The derivative of [G2]T [G1 ]T {N} with respect to {a} can be easily obtained using Eqs. 2.52,
2.53 and the chain rule. The element tangent stiffness matrix [k] can now be expressed as

[k] = [G2 ]T [G1 ]T [A][G1][G2 ] + [Bdkt ]T [D][Bdkt] + [Bdkt]T [B][G1][G2]
C1
⎡ ⎤
N̂ 0 0
⎢ ⎥
T T
+ [G2 ] [G1 ] [B][Bdkt] + [G2 ] T ⎢ 0 N̂ 0 ⎥ [G2 ] dA (2.54)
⎣ ⎦
0 0 N̂
The Newton-Raphson method using load control and arc-length control is illustrated first
for a concentrated load and then the steps that are to be modified for thermal and pressure
loads are shown. The arc-length method used in the present study is that of Crisfield [78].
2.3.2 Newton-Raphson method with load control

for a concentrated load
Let a known equilibrium configuration exist at some instant t = t1, the total load applied to
arrive at this configuration be f1 = λ1 qref and a1 be the solution at this instant. Here qref
is a fixed reference load and λ1 is the load proportionality factor. Since the reference load
qref considered here is a concentrated load in the global Cartesian coordinate system, it is
obtained just by prescribing the values of the global nodal forces. The new configuration at
some instant t = t2 is obtained by an incremental-iterative process in which a linearized form
of the equilibrium equations is solved in each step. Let Δf = Δλ qref be the incremental
load applied to the structure at equilibrium at instant t = t1 and a be the incremental
displacement vector. In each step the external load applied to the structure is held constant.
The incremental load factor Δλ can be held constant for all the steps or varied for each step.
The equations of equilibrium at t = t1 + Δt can be written as
gt1 +Δt = gt1 + [K] δa − Δλ qref = 0 (2.55)
Since the configuration at the beginning of the load increment Δλ is in equilibrium the
residue gt1 is identically zero. The solution to this incremental step can be obtained from
[K] δa = Δλ qref and the solution is updated as a = a + δa and a2 = a1 + a. The
incremental displacement a at the beginning of the step is obviously zero. The solution
a2 thus obtained is only an estimate of the actual solution, as the new configuration may
not be in equilibrium with the external forces. The correction to the solution (obtained by
linearizing the equations of equilibrium about the new solution point) is given by [K] δa =
−gt1 +Δt and the solution is updated as a = a + δa and a2 = a1 + a. This process is
repeated until the solution converges or the unbalance force diminishes to a very small
value. The residue gt1 +Δt is obtained by assembling the element residual vectors given by
Eq. 2.48 in the standard way. The force and moment resultants {N} and {M} and the
matrix [G1 ] used in Eq. 2.49 to compute the element internal force vector are evaluated at
each integration point using the local incremental displacement vector {a} which is obtained
from the global incremental displacement vector a in the standard way. The tangent stiffness
matrix [K] is updated every time the solution is updated (regular Newton-Raphson method).
If the tangent stiffness matrix is computed only at the beginning of each increment and not
updated after every iteration (modified Newton-Raphson method), especially in the case of
large rotation analysis, a large number of iterations will have to be performed or the solution
might even start diverging.
Several convergence criteria are available in the existing literature to check the
convergence of the solution during the incremental-iterative process. In the present study
the iterations are performed until the ratio of maximum correction to the displacements
in an iteration to maximum incremental translation and the ratio of maximum correction
to the rotations in an iteration to the maximum incremental rotation are both less than
10−3 . In other words if the solution vectors a and δa are split into displacements and
rotations as a = at + ar and δa = δat + δar , then the convergence criteria can be expressed
as δat ∞ /at ∞ < 10−3 and δar ∞ /ar ∞ < 10−3 . At the end of a load step, instead of
updating the stresses as given by Eq. 2.21, the force and moment resultants {N} and {M}
at each integration point for each element are updated as given by Eqs. 2.31 and 2.32 and
are used as {N1} and {M1 } for the next step.
2.3.3 Newton-Raphson method with arc-length control for a

concentrated load
The major difference between load control and arc-length control is that in the case of load
control the applied load in any given step is held constant, whereas in the case of arc-length
control the applied load is allowed to vary in any given step. In other words, the load factor
λ does not remain constant in a step in the case of arc-length control. Since the load factor
is an unknown, an equation in addition to the equilibrium equations (Eq. 2.48) is required
to determine the change in the load factor in each step. This is achieved by imposing a
constraint aT a = Δl2, where l is a constant called the arc-length. Since the load does
not remain constant in any step, the total load at some time t1 + Δt can be written as
ft1 +Δt = ft1 + δλ qref . The equlibrium equations at time t1 + Δt can now be expressed as
gt1 +Δt = gt1 + [K] δa − δλ qref = 0 (2.56)
Let Δλ0 be the constant incremental load parameter used for the incremental step as in
the case of load control. The solution to this incremental step is obtained as before from
[K] δa = Δλ0 qref and the solution is updated as a = a + δa and a2 = a1 + a. In the
subsequent iterations the correction to the solution (obtained by linearizing the equations of
equilibrium at the new solution point) is obtained from
δa = −[K]−1 gt1 +Δt + δλ [K]−1 qref = δā + δλ δat (2.57)
where the residue gt1 +Δt is obtained as in the case of load control and the change in the load
parameter δλ is obtained by applying the constraint as follows:
(a + δa)T (a + δa) = Δl2 (2.58)
where a is the solution at the end of the previous iteration (or at the end of the incremental
step) and a + δa is the solution that will be obtained at the end of the current iteration after
determining the change in the load parameter δλ. In Eq. 2.57 the updated value of [K] is
used. When δa from Eq. 2.57 is substituted into the constraint equation (Eq. 2.58), the
resulting quadratic equation is given by
a1 δλ2 + a2 δλ + a3 = 0 (2.59)
where
a1 = δaT
t δat
a2 = 2(a + δā)T δat

a3 = (a + δā)T (a + δā) (2.60)
Out of the two roots of Eq. 2.59, the one which results in the least root mean square error
between the solution at the end of the previous iteration a and the solution at the end of
the current iteration a + δa, is selected. The load factor should also be updated at the end
of each iteration in order to correctly compute the total load acting on the structure at any
time. For example, at the end of the first iterative step, the total load acting on the structure
will be (λ1 + Δλ0 + δλ)qref , or the load increment at the end of the first iteration will be
Δλ = Δλ0 + δλ and the total load acting on the structure will be f1 + Δλqref . The solution
at the end of the current iteration is updated as before as a = a + δa and a2 = a1 + a.
In the very first step, the incremental load parameter Δλ0 is suitably chosen and the arc
length Δl is determined as

Δl = Δλ0 (aT0 a0 ) (2.61)
where a0 is obtained from

[K] a0 = qref (2.62)
where [K] is the global tangent stiffness matrix at the beginning of the incremental step or at
time t = t1 and is made up of only the linear stiffness matrix and the initial stress matrix (the
last term in Eq. (2.54)). All the non-linear terms are zero since the incremental displacement
vector at the beginning of the load step is identically equal to zero. For subsequent steps,
the arc length may be adjusted such that a more or less constant number of iterations is
required to obtain convergence, using the following relation (Crisfield [78]):
Δli = Δli−1Nd /Ni−1 (2.63)
where Δli is the constant arc length to be used in the current step, Δli−1 is the constant
arc length used in the previous step, Ni−1 is the number of iterations needed in the previous
step and Nd is the desired number of iterations to be performed in any step to obtain
convergence based on the chosen convergence criterion. For all the steps other than the first,
the incremental loading parameter Δλ0 used for the tangential step is obtained by

Δλ0 = ±Δli/ (aTo a0 ) (2.64)
The sign follows that of the previous step (positive in the very first step) unless the
determinant of the tangent stiffness matrix has changed sign, in which case a sign reversal is
applied. In the present study the system of equations resulting from the linearization of the
governing equations of equilibrium (Eq. 2.50) is solved by factorizing the tangent stiffness
matrix in the standard way as [K] = [L][D][L]T , where [L] is a lower triangular matrix
with diagonal terms as unity and [D] is a diagonal matrix, the determinant of which gives
the determinant of [K]. In most practical problems the determinant turns out to be a very
large number which may not be represented using the finite precision of a computer. Hence
instead of computing the determinant of [K], the sign change is determined by determining
the number of negative terms in [D]. The determinant is exactly zero at the critical point
which could either be a limit point or a bifurcation point. Since the tangent stiffness matrix
is singular at the critical point, the solution at the critical point cannot be determined. Based
on the author’s experience, in most cases the critical point is not exactly reached. When the
arc length method is used, the critical point is usually crossed, resulting in a non-positive
definite tangent stiffness matrix. In the case of a limit point this does not cause any problems
as the solution at the limit point is unique and hence the load deflection path beyond the
limit point can be traced by changing the sign of the load parameter as mentioned above,
whereas in the case of a bifurcation point the solution is not unique and hence an additional
investigation is necessary.
Unlike the case of the limit point, the bifurcation point will have to determined a priori.
This can be done by monitoring the fundamental frequency of the tangent stiffness matrix.
The plot of the fundamental frequency versus the load parameter crosses the load-deflection
curve (plot of the deflection at any chosen point on the structure versus the load parameter)
at the critical point. Also, in the load-frequency diagram, the load for vanishing frequency
defines the critical point. Hence the critical point (limit or bifurcation point) can be
anticipated at an earlier stage. After arriving close to the critical point, the solution can be
perturbed by adding a small fraction of the mode shape corresponding to the fundamental
frequency and then proceed to determine if the secondary branches of the load-deflection
path are traced. An easier method would be to add small imperfections to the structure
right from the beginning and trace the load-deflection path. Since the load-deflection path
of the imperfect structure is asymptotic to the corresponding branch of the perfect structure
(depending on the type of the imperfection added), the response of the imperfect structure
gives a very good estimate of the secondary branches. The analysis is also less expensive,
unlike the analysis of a perfect structure, in which the eigenvalue analysis will have to be
performed at almost every load step to determine the critical point.
2.3.4 Newton-Raphson method for a thermal load

In the case of a thermal load, the load vector cannot be expressed as λ qref since a thermal
load cannot be treated as a proportional load. The Newton-Raphson method for a thermal
load is illustrated below. Only the steps required to compute the thermal load vector and
the residue are given below since all other steps remain the same.
As in the case of a concentrated load, a known equilibrium configuration is assumed to
exist at some instant t = t1 . Let ΔT = Δλ0 Tref be the temperature increment applied to
the structure, where Tref is the reference temperature field which at any point (ξ, η, z) within
the element is given by

1 z
Tref (ξ, η, z) = Tu(ξ, η) + Tl(ξ, η) + Tu(ξ, η) − Tl (ξ, η) (2.65)
2 h
where Tu(ξ, η) and Tl(ξ, η) are the temperatures at the top and bottom surfaces of the
laminate, respectively. They are represented in terms of the nodal quantities Tu1 , Tu2 , Tu3
and Tl1 , Tl2 , Tl3 using linear interpolation functions in area coordinates as follows:
Tu (ξ, η) = Tu1 + Tu21 ξ + Tu31 η Tl (ξ, η) = Tl1 + Tl21 ξ + Tl31 η (2.66)
where Tu21 = Tu2 − Tu1 , Tu31 = Tu3 − Tu1 , Tl21 = Tl2 − Tl1 , Tl31 = Tl3 − Tl1 .
The incremental force and moment resultants due to the applied temperature increment
ΔT are {ΔN 0} and {ΔM 0} as given by Eq. 2.36 ( {ΔN m } and {ΔM m } are zero since the
incremental displacement vector a at the beginning of the step is identically zero). The
element force vector due to the temperature increment is given by

0 0 0
{Δf } = [G2 ] [G1] {ΔN } + [Bdkt] {ΔM } dA
T T T
(2.67)
C1
The matrix [G1 ] in Eq. 2.67 consists of only zeroes and ones as the incremental displacement
a at t = t1 is identically zero. The element thermal load vectors {Δf 0} are assembled in a
standard way to obtain the global incremental thermal load vector Δf 0 . The solution to this
incremental step is obtained from [K] δa = Δf 0 and the solution is updated as a = a + δa
and a2 = a1 + a.
In the case of load control the correction to this solution is obtained as before using the
residue as [K] δa = −gt01 +Δt . The residue gt01 +Δt is obtained by assembling the element
residual vectors {g 0 } given by

0
{g } = [G2 ] [G1 ] {N} + [Bdkt ] {M} dA
T T T
(2.68)
C1
where {N} and {M} are given by Eqs. 2.31 and 2.32. The solution is updated as before as
a = a + δa and a2 = a1 + a. The thermal load vector is not used in evaluating the residue
as the thermal force and moment resultants are already contained in the expression for {N}
and {M} in Eqs. 2.31 and 2.32.
In the case of arc-length method the correction is obtained as
δa = −[K]−1 gt01 +Δt + δλ [K]−1 Δf 0 (2.69)
where δλ is determined from the constraint as in the case of a concentrated load and
Δf 0 is obtained by assembling the element thermal load vectors {Δf 0} given by Eq. 2.67.
The solution at the end of the current iteration is updated as before as a = a + δa and
a2 = a1 + a. The iterations are continued untill the solution converges based on the
convergence criterion as mentioned before. In determining the thermal load vector to be
used for the corrections δa (Eq. 2.69), the thermal force and moment resultants {ΔN 0 } and
{ΔM 0 } are computed using Eqs. 2.36 with ΔT = Tref , unlike the case of the incremental
step where the thermal force and moment resultants are determined using ΔT = Δλ0 Tref .
It should also be noted that the matrix [G1 ] that will be used in computing the thermal
load used in Eq. 2.69 will be a function of the displacements a as given by 2.39. Since the
thermal load depends on the displacements, it contributes to the tangent stiffness matrix.
By comparing Eqs. 2.49 and 2.67, it is obvious that the contribution to the tangent stiffness
matrix due to the variation of the thermal load with displacement is similar to the initial
stress matrix, which is the last term in Eq. 2.54. No additional effort is needed to determine
this contribution as the thermal force resultants are already included in the expressions for
{N} and {M} in Eqs. 2.31 and 2.32.
2.3.5 Newton-Raphson method for a pressure load

The pressure load is always assumed to act normal to the current or deformed configuration of
the structure. Hence the direction of the pressure load continuously changes as the structure
deforms and this is referred to as the follower action. Since the pressure load is a function
of the displacements, it contributes to the tangent stiffness matrix and this contribution,
called the pressure stiffness matrix, is in general unsymmetric. In practice a symmetric
form defined as the average of the pressure stiffness and its transpose is employed. The
contribution to the tangent stiffness matrix due to the follower effect, called the pressure
stiffness should be taken into account, especially in the case of large deformation analysis.
Neglecting the pressure stiffness might seriously affect the convergence of the solution even
when the deformations are small and in the case of large deformations the solution may not
converge at all.
Unlike the case of the thermal load, the determination of the pressure stiffness is not
straightforward. The pressure load and the pressure stiffness can be systematically derived
from the principle of virtual work. The virtual work due to the pressure load in a local
coordinate system is given by

δWp = {δu δv δw} P dĀ (2.70)
where δu, δv and δw are the virtual displacements of the midplane of the element, dĀ is the
area vector normal to the deformed configuration and P is the total pressure acting normal
to the element in the deformed configuration and is represented using a linear interpolation
in area coordinates as
P (ξ, η) = λ (P1ref (1 − ξ − η) + P2ref ξ + P3ref η) (2.71)
where P1ref , P2ref and P3ref are the nodal values obtained from a reference pressure
distribution to which the structure is subjected to and λ is the load proportionality factor.
The pressure distribution could either be space attached in which the distribution of the
magnitude is a function of the coordinates in the deformed configuration, or body attached
in which it is a function of the coordinates in the undeformed initial configuration. A positive
sign is used for the nodal pressure if the pressure acts in the same direction as the outward
normal to the current/deformed surface (internal pressure) and a negative sign is used for the
nodal pressure if the pressure acts opposite to the outward normal to the current/deformed
surface (external pressure).
In the updated Lagrangian formulation, all quantities are measured with respect to the
reference configuration C1 which is a known equilibrium configuration. Hence the area vector
in the deformed configuration dĀ has to be expressed in the reference configuration. This is
achieved using the relation given by Malvern [71] as
dĀ = [F −1]T {n̄} dA (2.72)
where [F ] is the deformation gradient tensor, {n̄} is the normal to the midplane of the
element in C1 and dA is the magnitude of an area element in C1 . The deformation gradient
tensor is given by ⎡ ⎤
1 + Ux Uy Uz
⎢ ⎥
[F ] = ⎢
⎣ Vx 1 + Vy Vz ⎥
⎦ (2.73)
Wx Wy 1 + Wz
where
U(x, y, z) = u(x, y) + zβx(x, y)

V (x, y, z) = v(x, y) + zβy (x, y)
W (x, y, z) = w(x, y) (2.74)
where u, v and w are the displacements of the mid-plane of the element along the local x, y
and z axes, respectively, and βx, βy are the rotations of the shell normal about the local y
and x axes, respectively.
Since the structures under consideration are very thin and most of the deformation of the
element is comprised of rigid body rotations, the contribution to the deformation gradient
tensor by the normal rotations βx and βy can be neglected and a linear interpolation for u, v
and w can be used. The deformation gradient tensor can now be simplified as
⎡ ⎤
1 + u,x u,y 0
⎢ ⎥
[F ] = ⎢
⎣ v,x 1 + v,y 0⎥
⎦ (2.75)
w,x w,y 1
Since the normal to the mid-plane of the element in C1 coincides with the local z-axis,
{n̄} = {0, 0, 1}. The area vector normal to the deformed configuration can now be expressed
in the reference configuration as
⎧ ⎫
⎪
⎪ v,x w,y − w,x (1 + v,y ) ⎪
⎪
⎨ ⎬
dĀ = ⎪ u,y w,x − w,y (1 + u,x) dA (2.76)
⎪ ⎪
⎪
⎩ ⎭
(1 + u,x)(1 + v,y ) − u,y v,x
The mid-plane displacements u, v and w can be expressed in terms of the nodal values using
linear interpolation functions in area coordinates as
u = (1 − ξ − η)u1 + ξu2 + ηu3

v = (1 − ξ − η)v1 + ξv2 + ηv3
w = (1 − ξ − η)w1 + ξw2 + ηw3 (2.77)
The variation of the mid-plane displacements can be expressed as
{δu δv δw}T = [G3 ][T ]{δa} (2.78)
where ⎡ ⎤
1−ξ−η 0 0 ξ 0 0 η 0 0
⎢ ⎥
[G3 ] = ⎢
⎣ 0 1−ξ −η 0 0 ξ 0 0 η 0⎥
⎦ (2.79)
0 0 1−ξ−η 0 0 ξ 0 0 η
The transformation matrix [T ] which is used to express the vector of nodal displacements
{u1, v1 , w1, u2, v2, w2 , u3, v3, w3 } in terms of the vector of nodal degrees of freedom of the shell
element {a} is comprised of zeroes and ones, the determination of which is straightforward.
The virtual work due to the pressure load can now be expressed as
δWp = {δa}T {fp } (2.80)
where {fp } is the pressure load vector in the local coordinate system given by
⎧ ⎫
⎪
⎪ v,xw,y − w,x (1 + v,y ) ⎪
⎪
1 1−η ⎨ ⎬
{fp } = [T ]T [G3 ]T P ⎪ u,y w,x − w,y (1 + u,x ) 2A dξdη (2.81)
0 − ⎪ ⎪
⎪
⎩ ⎭
(1 + u,x )(1 + v,y ) − u,y v,x
where A is the area of the mid-plane of the element in C1 . The partial derivatives of the
mid-plane displacements in Eq. 2.81, which are constants, are easily obtained from the linear
interpolation field as
1
{ u,x u,y v,x v,y w,x w,y }T = [G4 ][T ]{a} (2.82)
2A
where ⎡ ⎤
−b2 − b3 0 0 b2 0 0 b3 0 0
⎢ ⎥
⎢ −c2 − c3 0 0 c2 0 0 c3 0 0⎥
⎢ ⎥
⎢ ⎥
⎢ 0 −b2 − b3 0 0 b2 0 0 b3 0⎥
⎢ ⎥
[G4] = ⎢ ⎥ (2.83)
⎢ 0 −c2 − c3 0 0 c2 0 0 c3 0⎥
⎢ ⎥
⎢ ⎥
⎢ 0 0 −b2 − b3 0 0 b2 0 0 b3 ⎥
⎣ ⎦
0 0 −c2 − c3 0 0 c2 0 0 c3
where bi = yj − yk , ci = xk − xj and i, j, k take cyclically values 1,2,3. The pressure stiffness
[kp ] is obtained by taking the derivative of {fp } with respect to {a} using the standard chain
rule δ( ) = (∂( )/∂{a})δ{a} or in other words
{δfp} = (∂{fp }/∂{a}){δa} = [kp]{δa} (2.84)

After straightforward manipulations the pressure stiffness matrix can be expressed as

1 1−η
[kp ] = [T ]T [G3 ]T P [G5 ][G4 ][T ] dξdη (2.85)
0 0
where ⎡ ⎤
0 0 w,y −w,x −(1 + v,y ) v,x
⎢ ⎥
[G5] = ⎢
⎣ −w,y w,x 0 0 u,y −(1 + u,x) ⎥
⎦ (2.86)
1 + v,y −v,x −u,y 1 + u,x 0 0
The area integrals in Eqs. 2.81 and 2.85 are performed using the standard three-point
rule in area coordinates. As can be seen from Eq. 2.85, the pressure stiffness matrix is
unsymmetric. In the present study the pressure stiffness matrix is not split into interior
and boundary parts. A symmetric form defined as (1/2)([kp ] + [kp ]T ) is employed. The
pressure load f p and the pressure stiffness [Kp] in global Cartesian coordinates are obtained
by assembling the element load vectors given by Eq. 2.81 and the element stiffness matrix
given by Eq. 2.85 in the standard way. The pressure stiffness matrix is subtracted from the
tangent stiffness matrix (Eq. 2.54).
As in the case of mechanical load, a known equilibrium configuration is assumed to exist
at some instant t = t1. Let λ1 be the load proportionality factor at t = t1 and Δλ0 be
the incremental load proportionality factor at t = t1 . The incremental pressure load Δf p
acting on the current configuration (which at time t = t1 is C1 ) is obtained by assembling
the element pressure load vectors {f p } given by Eq. 2.81, using the load proportionality
factor λ = Δλ0 in Eq. 2.71. The solution to this incremental step is obtained as before from
[K] δa = Δf p and the solution is updated as a = a + δa and a2 = a1 + a. It should be noted
that for the incremental step the pressure stiffness is not included as the pressure load is not
a function of displacement (since a = 0 at t = t1 ).
In the case of load control the correction to this solution is obtained as before using the
residue as [K] δa = −gtp1+Δt . Here the pressure stiffness matrix is included in [K]. The
residue gtp1 +Δt is obtained by assembling the element residual vectors {g p } given by
{g p } = {q} − {f p } (2.87)
where {q} is the element internal force vector given by Eq. 2.49 (considering only mechanical
stresses) and {f p } is given by Eq. 2.81. In computing {f p }, the current load proportionality
factor is used in Eq. 2.71, which is λ1 + Δλ0. The solution is updated as before as a = a + δa
and a2 = a1 + a.
In the case of the arc length method, the correction is obtained as
δa = −[K]−1 gtp1 +Δt + δλ [K]−1 Δf p (2.88)
where δλ is determined from the constraint as in the case of a concentrated load. The solution
at the end of the current iteration is updated as before as a = u + δu and a2 = a1 + u. The
iterations are continued until the solution converges based on the convergence criterion as
mentioned before. The pressure load Δf p in Eq. 2.88 is obtained using λ = 1 in Eq. 2.71
unlike the case of the incremental step where λ = Δλ0 . The pressure load used in the residue
(Eq. 2.87) is determined using the current load proportionality factor λ1 + Δλ0 at the end
of the incremental step or λ1 + Δλ0 + δλ at the end of an iteration.
2.3.6 Dynamic analysis

As in the case of static analysis, it is assumed that a known equilibrium configuration exists
at some time t1 and the solution at some time t1 +Δt has to be determined. The equations of
equilibrium governing the dynamic response of a structure at time t1 + Δt can be expressed
as
[M]ät1 +Δt + qt1+Δt = ft1 +Δt (2.89)
where M is the mass matrix, q is the internal force and f is the externally applied force.
Since the transverse displacement is not defined explicitly at the interior of the DKT
element, the determination of a consistent mass matrix is not possible. This problem is
alleviated by using a lumped mass matrix which is obtained by assigning one-third of the
total mass of the element to each of the translational degrees of freedom at each node. The
rotational inertias are neglected in computing the lumped mass matrix. It is well known
that the natural frequencies obtained using a lumped mass converge from below. A fine
mesh is required in order to obtain the the eigenvalues of a large number of modes and
hence the analysis is computationally expensive. In the present study a so called consistent
mass matrix is also employed in order to compare the performance of the two types of mass
matrices. A 9x9 mass matrix corresponding to the degrees of freedom of the DKT element
is obtained using the shape functions in area coordinates proposed by Specht [37] for the
transverse displacement. A 12x12 mass matrix corresponding to the degrees of freedom of
the LST element is obtained using the same shape functions that are used for the in-plane
displacements in the LST element. The 9x9 mass matrix corresponding to the degrees of
freedom of the membrane element is obtained by applying the transformation given by Cook
[5] to the 12x12 mass matrix of the LST element derived as mentioned above. A three
point integration scheme in area coordinates (Yang [2]) is used to determine the elements
of the mass matrix. The element mass matrices thus obtained in a local coordinate system
are transformed to the global Cartesian coordinate system prior to their assembly into the
global mass matrix [M]. Details of derivation of the mass matrices are not repoduced here
as it is well documented in the existing literature. The frequencies of free vibration and the
mode shapes are obtained by solving the eigenvalue problem given by

2
[K] − ω [M] =0 (2.90)
where [K] is the linear stiffness matrix obtained from Eq. 2.54 by neglecting the initial stress
matrix and the terms dependent on displacements. The eigenvalue problem given by Eq. 2.90
is solved using the subspace iteration method, which is described in detail by Bathe [79]. The
non-linear governing equations of dynamic equilibrium given by Eq. 2.89 are solved using
the Newmark time integration method in conjunction with the Newton-Raphson method,
using the algorithm given by Bathe and Ramm [80].
Chapter 3
Numerical examples
A large set of problems on linear static analysis, free vibration analysis, linear thermal
analysis and geometrically nonlinear analysis was solved in order to demonstrate the accuracy
of the present flat shell formulation, and the results are presented in this chapter in different
sections. The present results are compared with those available in the existing literature
whereever applicable and with those obtained using the STRI3, S4R and S4R5 elements
of the commercial finite element package ABAQUS using the same solution procedure and
convergence criterion. STRI3 is a three-noded flat shell element obtained by combining the
DKT plate bending element and the CST element. S4R is a four-noded doubly curved shell
element with reduced integration, applicable for both thick and thin shell. S4R5 is also a
four-noded double curved shell element with reduced integration, but is applicable only for
thin shells.
3.1 Linear Static Analysis

Static analysis of simply supported square plates under a doubly sinusoidal load
P0 sin(πx/a)sin(πy/a) and simply supported spherical shells with square base under a
uniformly distributed load of intensity P0 were performed. Both symmetric and unsymmetric
laminates were considered for the analysis. In all cases the layers constituting the laminate
were assumed to be of equal thickness. Unless mentioned otherwise, for the square plates
h/a = 0.01 where a is the side and h is the thickness of the plate, and for the spherical
shells h/a = 0.01, r/a = 10 where a is the side of the square base, h is the thickness and r is
51
CHAPTER 3. NUMERICAL EXAMPLES 52
the radius of curvature. The following types of boundary conditions and material properties
were used for the numerical examples presented.
BC1: v = w = θx = 0 at x = 0, a and u = w = θy = 0 at y = 0, b
BC2: u = w = θx = 0 at x = 0, a and v = w = θy = 0 at y = 0, b
MAT1: E1 = 25E2 , G12 = 0.5E2 , ν12 = 0.25
MAT2: E1 = 40E2 , G12 = 0.6E2 , ν12 = 0.25
Simply supported [0/90/90/0] square plate: One quadrant of the plate under a doubly
sinusoidal load was analyzed with boundary conditions BC1 and material properties MAT1.
The convergence of non-dimensionalized mid-deflection is given in Table 3.1 along with the
closed form solution as given by Phan and Reddy [81].
Simply supported [-45/45] square plate: A full plate under a doubly sinusoidal load
was analyzed with boundary conditions BC2 and material properties MAT2. The
convergence of non-dimensionalized maximum displacements w∗ = wmax E2 h3 102 /Po a4,
u∗ = umax E2 h3 104 /Po a4 and v∗ = vmaxE2 h3 104 /Po a4 is given in Table 3.2 along with the
analytical solution computed using equations 39-44 of Whitney and Leissa [82].
Scordelis-Lo roof: The geometry, material properties and the boundary conditions used are
shown in Fig. 3.1. One quadrant of the roof under selfweight was analyzed. The convergence
of the vertical downward deflection at the mid-point of the free edge is given in Table 3.3 along
with the results of Chen [59] (DKT + 4 point integrated higher order membrane element
of Allman [4] ), Carpenter et al. [10] (DKT + one point integrated quadratic membrane
element of Allman [3]) and those obtained using STRI3. The exact value of the vertical
downward deflection at the mid-point of the free edge as quoted by Carpenter et al. [10] is
0.3024.
Pinched cylinder: The geometry, material properties and the boundary conditions used are
shown in Fig. 3.2. One quadrant of the cylinder under a concentrated mid-load (as shown
in Fig. 3.2) was analyzed. The convergence of the vertical downward deflection at the load
point is given in Table 3.4 along with the results of Chen [59], Carpenter et al. [10] and those
obtained using STRI3. The exact value of the vertical downward downward deflection at the
load point as quoted by Carpenter et al. [10] is 1.8248 x 10−5 .
Hemispherical shell with a hole: The geometry, material properties and the boundary
conditions used are shown in Fig. 3.3. One quadrant of the cylinder under diametrically
opposite pinching and stretching loads (as shown in Fig. 3.3) was analyzed. The boundary
conditions as shown in Fig. 3.3 (and as given in the literature) do not inhibit the rigid body
vertical translation of the shell. Analysis was performed both with a free hole and with the
vertical displacement constrained along the hole. The convergence of the radial deflection
at A, in the direction of the load, and the vertical deflection at A, are given in Tables 3.5
and 3.6 respectively, for the case with a free hole, along with the results of Chen [59] and
those obtained using the STRI3 element of ABAQUS. The corresponding results for the case
with the vertical displacement constrained along the hole are given in Tables 3.7 and 3.8.
The results of Carpenter et al. [10] are not compared since their results correspond to a
hemispherical shell without a hole. The exact value of the radial deflection under the load
for the shell without a hole as quoted by MacNeal and Harder [83] is 0.0924. To the best
of the author’s knowledge the exact solution for the shell with a hole is not available in the
existing literature.
Simply supported spherical shell: The full shell under a uniformly distributed load was
analyzed with boundary conditions BC1 and material properties MAT1. The convergence
of non-dimensionalized mid deflection for two stacking sequences, [0/90] and [0/90/90/0] is
given in Table 3.9 along with the exact solution as given by Reddy [84].
The results for linear static analysis converged from below and to within 1 % of the
exact solution or finite element solution available in the existing literature. The three shell
problems, namely, Scordelis-Lo roof, pinched cylinder and the hemispherical shell have been
widely used in the existing literature as standard problems to test the accuracy of the finite
element formulations. Hence a discussion on the performance of the present formulation in
the case of these three problems is warranted here.
In the case of the Scordelis-Lo roof, the convergence of the present results is less rapid
compared to that of the underintegrated Allman element of Carpenter et al. [10] and the
higher order Allman element of Chen [59], but the same accuracy is obtained using a finer
mesh without the need for the additional effort that is required to suppress the spurious zero
energy modes as in the case of the underintegrated element. Since the Allman element used in
the present formulation is quadratic, the present analysis is expected to be computationally
less expensive compared to that using the higher order Allman element, especially in the
case of non-linear analysis.

In the case of the pinched cylinder, the present results appear to converge towards the
exact solution, whereas both the underintegrated Allman element used by Carpenter et al.
[10] and the higher order Allman element used by Chen [59] give over soft solutions. The
STRI3 element appeared to converge towards the exact solution less rapidly compared to
the present results.
It has been well established in the existing literature that the fully integrated Allman
element (quadratic membrane element of Allman [3]) exhibits very poor rate convergence
and that the CST element outperforms the fully integrated Allman element in the case of
the hemispherical shell under diametrically opposite pinching and stretching loads. This can
be clearly seen from the results presented in Table 3.5. The 4-point integrated higher order
membrane element used by Chen [59] performs better than the present element, but the
convergence is not as rapid as in the case of the CST element.
The boundary conditions for this problem as given by MacNeal and Harder [83] and Chen
[59] do not suppress the rigid body vertical translation of the entire shell. As can be seen from
Table 3.6, the vertical deflection at point A does not show any sign of convergence. This can
be attributed mainly to the fact that the rigid body motion of the shell is not completely
restrained. Though the stiffness matrix may not be singular, it could be ill conditioned,
giving incorrect results. When the vertical deflection along the hole was suppressed, the
present results and those obtained using STRI3 and S4R converged to more or less the same
value for both the radial and vertical deflection at point A, but the rate of convergence of the
present results is less rapid compared to that of STRI3. The reason for better performance
of the CST element for this problem alone is not quite clear.
3.2 Free Vibration Analysis

Free vibration analysis of square cantilever plates, skew cantilever plates and simply
supported spherical shells was performed. Both symmetric and unsymmetric laminates were
considered for analysis. In all cases layers constituting the laminates were assumed to be of
equal thickness.
Eight layer [0/45/-45/90/90/-45/45/0] cantilever plate: The material properties used are:
E1 = 128 GP a, E2 = 11 GP a, ν12 = 0.25, G12 = 4.48 GP a, density = 1500 kg/m3 . Plate
dimensions are 0.0762 m x 0.0762 m x 0.00105664 m. The convergence of the first five
natural frequencies (Hz) are given in Tables 3.10 and 3.11 for lumped and consistent mass
matrices respectively, along with the Ritz solution obtained by Kapania and Singhvi [85].
The mode shapes obtained using the consistent mass matrix are shown in Fig 3.5.
Two-layer [0/22.5] square cantilever plate: The material properties used are: E1 =
161.21 GP a, E2 = 12.52 GP a, ν12 = 0.22, G12 = 6.75 GP a, density = 1881.81 kg/m3 .
Plate dimensions are 0.2032 m x 0.2032 m x 0.00124968 m. The convergence of the first five
natural frequencies (Hz) is given in Tables 3.12 and 3.13 for lumped and consistent mass
matrices, respectively, along with the Ritz solution obtained by Kapania and Singhvi [85].
The mode shapes obtained using the consistent mass matrix are shown in Fig. 3.6.
Six layer [152 /0]s skew cantilever plate : The geometry of the skew cantilever plate is shown
in Fig. 3.4. The material properties used are: E1 = 38.61 GP a, E2 = 8.27 GP a, ν = 0.26,
G12 = 4.14 GP a, density=2546.54 kg/m3 . The geometric parameters are aspect ratio=2.333,
taper ratio=0.5, area=0.04064508 m2, thickness=0.003556 m. The sweep angle β of the
quarterchord line as shown in Fig. 3.4 is 30o . The aspect ratio is defined as the ratio of
length to the width at the root. The taper ratio is defined as the ratio of width at the tip to
the width at the root. The fiber angle is measured counterclockwise from the quarterchord
line. For example a 2-ply laminate [0/45] has fibers along the quarterchord line in one layer
and at 45 degrees counter clock wise from the quarterchord line in the other layer. The
convergence of the first five natural frequencies (Hz) is given in Tables 3.14 and 3.15 for
lumped and consistent mass matrices, respectively, along with the Ritz solution obtained by
Kapania and Singhvi [85]. The mode shapes obtained using the consistent mass matrix are
shown in Fig. 3.7.
Two-layer [0/22.5] skew cantilever plate: The material properties used are, E1 =
161.21 GP a, E2 = 12.52 GP a, ν = 0.22, G12 = 6.75 GP a, density=1881.81 kg/m3 . The
geometric parameters are β = 45, aspect ratio=1, taper ratio=1, area=0.04129024 m2,
thickness=0.00124968 m. The convergence of the five natural frequencies (Hz) are given in
Tables 3.16 and 3.17 for lumped and consistent mass matrices respectively, along with the
Ritz solution obtained by Kapania and Singhvi [85]. The mode shapes obtained using the
consistent mass matrix are shown in Fig. 3.8.
Simply supported [0/90/90/0] spherical shell: The material and geometric properties used
are: E1 = 25E2 , G12 = 0.5E2 , ν12 = 0.25, a/h = 100, r/a = 10. The full shell was
analyzed with boundary conditions as in the static analysis. The convergence of first five

non-dimensional frequencies (ωa2 ρ/E2 /h) is shown in Tables 3.18 and 3.19 for lumped
and consistent mass matrices respectively. The exact value for the non-dimensionalized
fundamental frequency as given by Reddy [84] is 20.38. The mode shapes obtained using
the consistent mass matrix are shown in Fig. 3.9.
The natural frequencies obtained using the lumped mass matrix converged from below
for the plate problems. No regular trend was observed in the case of shells and with the
results obtained using the consistent mass matrix. The rate of convergence of the frequencies
obtained using the consistent mass matrix was very rapid compared to that obtained using
the lumped mass matrix, except in the case of the spherical shells for which the rate of
convergence of the frequencies obtained by both approaches was more or less equal. The
results obtained using both approaches are in good agreement with each other and with
those of Kapania and Singhvi [85] except for the higher modes in the case of symmetrically
laminated square plates. There is a lot of disparity among the free vibration results by
different authors, as stated by Kapania and Singhvi [85]. Since exact solutions are not
available, nothing can be said about the accuracy of the results available in the existing
literature. However, the consistent mass approach adopted here seems far more economical
than the lumped mass approach, especially when a large number of modes is required to be
extracted.
3.3 Linear Thermal Analysis

Thermal analysis of square plates and spherical shells with a square base was performed.
Symmetric and unsymmetric laminates under a uniform thermal field and a non-uniform
thermal field with a linear variation of temperature across the thickness were considered. For
both uniform and non-uniform temperature distributions, the top surface of the laminate is
o
assumed to be at a constant temperature of Tu C and the bottom surface at a constant
o
temperature of Tl C. The geometric parameters used are h/a = 0.01 for square plates and
h/a = 0.01, r/a = 10 for spherical shells. The following boundary conditions and material
properties were used for the numerical problems presented:
BC1: v = w = θx = 0 at x = 0, a and u = w = θy = 0 at y = 0, b
BC2: u = w = θx = 0 at x = 0, a and v = w = θy = 0 at y = 0, b
Material properties: E1 = 53.8 GP a, E2 = 17.9 GP a, G12 = 8.62 GP a, ν12 = 0.25,
α1 = 6.3x10−6 /o C, α2 = 20.5x10−6 /o C, α12 = 0
The results obtained for simply supported, symmetric, specially orthotropic plates and
antisymmetric cross-ply and angle-ply laminated plates are compared with the series solution
presented by Wu and Tauchert [86] & [87]. A convergence study was first performed in order
to determine the number of terms to be used for the series solution. The convergence history
showed that the convergence of the in-plane displacements was very slow compared to the
rapid convergence of the transverse displacement. In the numerical examples presented
below, the mid-deflection wmax and the maximum in-plane displacements umax and vmax
obtained using the present finite element formulation are compared with the series solution
of Wu and Tauchert [86] & [87] obtained using 100 terms in the case of wmax of symmetric
specially orthotropic laminates and 10000 terms (100x100) in the case of umax and vmax
of symmetric specially orthotropic laminates and umax, vmax and wmax of antisymmetric
cross-ply and angle-ply laminates.
Simply supported [0/90/90/0] square plate (BC1): The convergence of wmax due to a
linearly varying thermal field (Tu = 1, Tl = −1) and umax, vmax due to a uniform thermal
field (Tu = 1, Tl = 1) are given in Table 3.20 along with the series solution of Wu and
Tauchert [86] & [87].
Simply supported [0/90/0/90] square plate (BC1): The convergence of umax , vmax and wmax
due to a non-uniform thermal field (Tu = 1, Tl = 0) are given in Table 3.21 along with the
series solution of Wu and Tauchert [86] & [87].
Simply supported [15/ − 15/15/ − 15] square plate (BC2): The convergence of umax , vmax and
wmax due to a non-uniform thermal field (Tu = 1, Tl = 0) are given in Table 3.22 along with
the series solution of Wu and Tauchert [86] & [87].
Simply supported spherical shell (BC1): The convergence of umax , vmax and wmax due to a
non-uniform thermal field (Tu = 1, Tl = 0) for cross-ply and angle-ply laminates is given in
Tables 3.23 and 3.24, respectively. The results could not be compared as exact solutions or
finite element results for thermal deformation of thin shells are not available in the existing
literature to the best of the author’s knowledge.

The thermal deformations converged from below in the case of plates. In the case of
shells the transverse deformations converged from above. No regular trend was noticed in
the case of in-plane displacements. The present results (in the case of plate problems) are
in good agreement with the results available in the existing literature.
3.4 Geometrically Nonlinear Analysis
3.4.1 Static response analysis

Two layer [45/-45] hinged cylindrical panel: The geometry, material properties and the
boundary conditions used are given in Fig. 3.10. One quarter of the panel under a
concentrated load of 1600 N was analyzed. The present results are compared with those
obtained using STRI3 and S4R elements in Fig. 3.11. The load -deflection curve was traced
using the standard arc-length control method. Very small increments have been used in
order to obtain a very smooth load-deflection curve. Only a few points are shown in the
figure. It was possible to perform the analysis using very large load steps since the rotations
are very small.
Cantilever beam under a tip moment : The geometry and material properties used are:
Length = 10 m, width = 1 m, thickness = 0.1 m, E = 1.2x108 N/m2 , ν = 0. The applied
bending moment (2000π Nm) at the tip causes the beam to deform into a full circle. In Fig.
3.12 the present results, obtained using load-control in 50 steps and a total of 200 iterations,
are compared with those obtained using the STRI3, S4R5 and S4R elements. The present
results are in good agreement with those obtained using STRI3 and S4R elements. The
results obtained using the S4R5 element show a slight deviation from the rest of the results
for 4200 < M < 6000 Nm.
Clamped cylindrical shell under pinching loads at the apex: The geometry and material
properties used are: Length = 3.048 m, Radius = 1.016 m, thickness = 0.03 m, E = 2.0685
x 107 N/m2 and ν = 0.3. One quarter of the shell, with the boundary conditions as shown
in Fig. 3.13, was analyzed under a concentrated apex load. The analysis was performed
until the deflection under the load was just about equal to the radius of the cylinder. The
load-deflection path was traced using arc length control in 32 increments with a total of 120
iterations. Though the response does not show any instabilities, arc length control has been
used because the convergence is faster in the case of arc length control than load control. In
Fig. 3.13 the present results are compared with those obtained using STRI3, S4R and those
of Brank et al. [57]. The STRI3 element showed severe convergence problem for P > 500 N
when load control was used. Though large load increments could be used up to P = 500
N, the load increments were automatically cut down for P > 500 N resulting in at least 100
steps. When arc length control was used, the STRI3 element showed extreme sensitivity to
the size of the arc-length increments. Only the results obtained using the arc-length control
are presented in the case of the STRI3 element. The flat shell element of the present study
and S4R did not show such problems. As can be seen from Fig. 3.13, the results obtained
using the STRI3 element in 37 increments do not agree well with the rest of the results. The
present results agree well with those of Brank et al. [57], S4R and STRI3 (obtained using 90
increments).
Cylindrical shell with free ends under stretching loads at the apex: The geometry and the
material properties used are: Radius = 4.953 m, Length = 10.35 m, thickness = 0.094 m, E
= 10.5 x 103 N/m2 and ν = 0.3125. One quarter of the shell, with the boundary conditions
as shown in Fig. 3.14, was analyzed under stretching loads at the apex. Two distinct zones
of stiff and soft behavior can be identified from the load deflection curve (3.14). Most of
the results in the existing literature show good agreement only in the soft zone. In 3.14 the
present results, obtained using arc-length control in 25 steps with a total of 95 iterations,
are compared with those of Brank et al. [57] and those obtained using the S4R and S4R5
elements of ABAQUS. The present results are in good agreement with those of Brank et al.
[57] and S4R in both the stiff and soft zones. The S4R5 element failed to proceed in the
same direction for P > 32 N.
Hemispherical shell with a hole under pinching and stretching loads: The geometry,
material properties and the boundary conditions used are the same as those used in the linear
analysis. The non-linear analysis was performed with the transverse deflection constrained
along the hole in order to prevent the rigid body motion which otherwise is not suppressed
by the symmetry conditions shown in Fig. 3.3. The present results obtained using the arc-
length control in 22 steps with a total of 88 iterations are compared with those of Buechter
and Ramm [88] and those obtained using STRI3 and S4R elements in Figures 3.15 and 3.16.
The present results are in good agreement with those of Buechter and Ramm [88] and S4R.
The results obtained using the STRI3 element for the transverse displacement Fig. 3.16 do
not agree well with the present results and those obtained using S4R.
Ring plate under a line load at the free edge: The geometry (Fig. 3.17) and material
properties used are: inner radius = 8 m, outer radius = 10 m, thickness = 0.03, E = 2.1 x
108 N/m2 and ν = 0.0. The plate is clamped at one of the edges which meet along the line
AB (Fig. 3.17) and subjected to a line load at the other edge. The present results obtained
using arc-length control in 44 steps and 223 iterations are compared with those of Buechter
and Ramm [88] and those obtained using the S4R and STRI3 elements in Fig. 3.17 and are
in good agreement. As in the case of the pinched cylindrical shell, the STRI3 element showed
some convergence problems. Though large initial increments were used (in arc-length control)
the increments were cut down several times during the course of the analysis resulting in
242 increments with a total of 1199 iterations. The total number of iterations used in the
case of the present formulation or S4R is smaller than the total number of increments used
in the case of the STRI3 elements.
3.4.2 Dynamic response analysis

The dynamic response analysis of an antisymmetric angle ply laminated ([45/-45]) hinged
cylindrical panel of thickness 0.0126 m under a centrally applied load of 1600N was
performed. The geometry and material properties used are the same as those used in the
static response analysis. The density and time step used are 1000 kg/m3 and 0.5x10−3 sec.
respectively. The force was applied at time t = 0 and held constant. The results obtained
using the present formulation are compared with those obtained using the STRI3 and S4R
element (using the same time step) in Fig. 3.18 and are in good agreement.
3.4.3 Thermal postbuckling analysis

The thermal postbuckling analysis of a simply supported symmetrically laminated imperfect
square plate and an antisymmetric angle ply laminated clamped cylindrical panel subject
to a uniform temperature rise were performed. The material properties used for the plate
are E1 = 155Gpa, E2 = 8.07Gpa, ν12 = 0.22 and G12 = 4.55Gpa. The plate dimensions
are 0.15x0.15x0.001016 m and the imperfections considered are in the form of the first
buckling mode. The bifurcation behavior of the perfect plate at the critical temperature was
simulated by adding a very small amount of imperfection, with a maximum magnitude of

0.0005 times the thickness of the plate. The critical temperature (39.491 0 C) was determined
by performing a linear thermal buckling analysis. The same shape functions that were
used to compute the so-called consistent mass matrix were used to compute the geometric
stiffness matrix for the linear buckling analysis. The eigenvalue problem was solved using a
subspace iteration method. The material properties used for the clamped cylindrical panel
are E1 = 138Gpa, E2 = 8.28Gpa, ν12 = 0.33 and G12 = 6.9Gpa. The geometry is the same
as that used in the static analysis except that the thickness used in this case is 0.00254 m.
The results for thermal post buckling analysis of the square plate are given in Fig. 3.19
along with the Rayleigh-Ritz solution of Meyers and Hyer [89] and are in good agreement.
The results given in Fig. 3.19 are nondimensionalized with respect to the critical buckling
temperature (38.6 o C) given by Meyers and Hyer [89]. The response of the imperfect plate
closely follows the simulated response of the perfect plate as in the case of the results
presented by Meyers and Hyer [89]. The results for the clamped cylinder under uniform
temperature rise are given in Fig. 3.20 along with those obtained using the STRI3 and S4R
elements. The present results are in good agreement with those obtained using the S4R
element, whereas the results obtained using the STRI3 element do not agree well with the
present results or those obtained using S4R.
3.4.4 Analysis of general shells under deformation-dependent

pressure loads
Very few numerical examples involving deformation dependent pressure loads are available
in the existing literature. To the best of the author’s knowledge, a cantilever beam under
uniform external pressure and a thin circular ring under a non-uniform external pressure are
the two most widely used examples. These two examples were solved in order to validate
the accuracy of the present formulation. Though the response in these two examples does
not show any instability (within the range of the applied loads), the analysis was performed
using the Riks method13 as this method converges faster than the load-control method.
Cantilever beam under uniform external pressure: The geometry and material properties
used are given in Fig. 3.21. The present results are compared in Fig. 3.21 with those obtained
using S4R and are in good agreement. In order to study the effect of the pressure stiffness
matrix on the convergence and accuracy of the solution, the analysis was performed both with
and without including the pressure stiffness matrix in the system tangent stiffness matrix.
The results of this comparative study are presented in Fig. 3.22, where Case-1 and Case-2
are with and without the pressure stiffness matrix, respectively. The results of Case-1 are
the same as those presented in Fig. 3.21, but have been presented again for comparison. As
can be seen from Fig. 3.22, the results in both the cases agree well, but in Case-2 very small
load steps were required to obtain convergence beyond a certain load level. The analysis
was abandoned beyond a certain load factor as the rate of convergence was extremely slow.
This indicates that exclusion of the pressure stiffness matrix could inflict significant harmful
effects on the convergence of the solution, especially in the regime of large deformations and
rotations.
Thin circular ring under non-uniform external pressure: The geometry and the material
properties used are given in Fig. 3.23. Since the geometry and the loading are symmetric, one
half of the ring was modeled using a 1x40 mesh of 80 elements. The pressure distribution was
assumed to be body attached, or in other words the magnitude of the pressure distribution
is a function of the coordinates in the undeformed initial configuration, but the direction
is always normal to the current/deformed configuration. Present results are compared with
the numerical solution of Seide [90] in Fig. 3.23 and are in good agreement. The pressure
stiffness matrix did not have a significant effect on the convergence of the solution in this
problem.
Table 3.1: Convergence of non-dimensionalized mid-

deflection of a simply supported [0/90/90/0] square
plate under doubly sinusoidal load.
Grid size d.o.f. wE2 h3102 /Po a4

10x10 726 0.429483
12x12 1014 0.430029
14x14 1350 0.430356
16x16 1734 0.430567
18x18 2166 0.430711
20x20 2646 0.430814
22x22 3174 0.430889
24x24 3750 0.430947
Phan and
Reddy(1985) 0.434300
Table 3.2: Convergence of non-dimensionalized

maximum displacements u∗, v∗ and w∗ of a simply
supported [-45/45] square plate under a doubly
sinusoidal load.
Grid size d.o.f. u∗ v∗ w∗

10x10 726 0.3396 0.3396 0.4512
12x12 1014 0.3424 0.3424 0.4560
14x14 1350 0.3441 0.3441 0.4588
16x16 1734 0.3452 0.3452 0.4607
18x18 2166 0.3459 0.3459 0.4620
20x20 2646 0.3465 0.3465 0.4629
22x22 3174 0.3469 0.3469 0.4635
24x24 3750 0.3472 0.3472 0.4641
Whitney and
Leissa (1969) 0.3486 0.3486 0.4667
Table 3.3: Convergence of the vertical downward deflection at the

mid-point of the free edge of the Scordelis roof.
Grid size 8x8 10x10 12x12 16x16 20x20

Present 0.2986 0.2991 0.2994 0.2998 0.3001
Chen (1992) 0.2999 0.2999 0.3001 0.3002 -
Carpenter el al. (1985 ) 0.3006 0.3003 0.3003 0.3003 0.3003
STRI3 0.2642 0.2752 0.2819 0.2896 0.2933
Table 3.4: Convergence of the vertical downward deflection (x103 ) at the

load point of a pinched cylinder.
Grid size 8x8 10x10 12x12 16x16 20x20 23x23

Present 1.6183 1.7060 1.7508 1.7906 1.8066 1.8130
Chen (1992) 1.7201 1.8051 1.8412 1.8600
Carpenter el al. (1985 ) 1.7353 1.8138 1.8485 1.8631 1.8613 1.8576
STRI3 1.5740 1.6770 1.7300 1.7770 1.7970 1.8050
Table 3.5: Convergence of the radial deflection in the direction of

the load, at point A, of a hemispherical shell with a free hole.
Grid size 8x8 10x10 12x12 16x16 20x20 23x23

Present 0.0354 0.0524 0.0664 0.0819 0.0881 0.0902
Chen (1992) 0.0759 0.0849 0.0888 0.0917
STRI3 0.0927 0.0926 0.0925 0.0924 0.0925 0.0926
Table 3.6: Convergence of the vertical deflection at the load

point A of a hemispherical shell with a free hole.
Grid size 10x10 14x14 18x18 24x24 28x28

Present +0.0212 -0.1045 +0.0246 +0.0756 +0.0536
STRI3 -0.1160 +0.0397 +0.0239 -0.0165 -0.9200
Table 3.7: Convergence of the radial deflection in the direction

of the load, at point A, of a hemispherical shell, with the
vertical deflection constrained along the hole.
Grid size 10x10 14x14 18x18 24x24 28x28 36x36

Present 0.0517 0.0735 0.0824 0.0870 0.0881 0.0889
STRI3 0.0882 0.0883 0.0884 0.0887 0.0888 0.0890
Table 3.8: Convergence of the downward vertical deflection

at the load point A, of a hemispherical shell, with the vertical
displacement constrained along the hole.
Grid size 10x10 14x14 18x18 24x24 28x28 36x36

Present 0.0256 0.0363 0.0406 0.0428 0.0432 0.0436
STRI3 0.0438 0.0436 0.0435 0.0435 0.0436 0.0437
Table 3.9: Convergence of non-dimensionalized

mid- deflection of simply supported [0/90] and
[0/90/90/0] spherical shells under a uniformly
distributed load.
Grid size d.o.f. wE2 h3 103/P0 a4

full shell [0/90] [0/90/90/0]
10x10 726 5.320 3.600
16x16 1734 5.447 3.663
20x20 2646 5.477 3.679
24x24 3750 5.494 3.687
28x28 5046 5.503 3.691
30x30 5766 5.507 3.693
32x32 6534 5.510 3.694
Reddy (1984) 5.542 3.720
Table 3.10: Convergence of first five natural frequencies (Hz) of

a square [0/45/-45/90/90/-45/45/0] cantilever plate using lumped
mass.
mesh mode 1 mode 2 mode 3 mode 4 mode 5

6x6 223.09 424.93 959.98 1364.01 1627.73
8x8 224.54 426.29 986.88 1391.17 1673.51
10x10 225.20 426.84 1000.08 1404.03 1694.57
12x12 225.55 427.13 1007.44 1411.10 1705.88
14x14 225.76 427.29 1011.93 1415.38 1712.64
16x16 225.90 427.38 1014.87 1418.18 1717.00
18x18 225.99 427.45 1016.89 1420.10 1719.96
20x20 226.05 427.49 1018.34 1421.47 1722.08
22x22 226.10 427.53 1019.41 1422.49 1723.63
24x24 226.14 427.55 1020.23 1423.26 1724.81
Kapania and
Singhvi(1991) 225.14 425.63 1020.21 1418.18 1721.59
Table 3.11: Convergence of first five natural frequencies (Hz) of a

square [0/45/-45/90/90/-45/45/0] cantilever plate using consistent
mass.

6x6 225.91 429.59 1040.47 1413.20 1728.79
8x8 226.13 428.89 1033.73 1419.74 1731.33
10x10 226.22 428.52 1030.54 1422.74 1732.03
12x12 226.26 428.29 1028.77 1424.26 1732.07
14x14 226.28 428.14 1027.68 1425.14 1731.96
16x16 226.30 428.04 1026.97 1425.69 1731.83
18x18 226.31 427.97 1026.47 1426.05 1731.70
20x20 226.31 427.91 1026.11 1426.31 1731.60
Kapania and
Singhvi(1991) 225.14 425.63 1020.21 1418.18 1721.59
Table 3.12: Convergence of the first five natural frequencies (Hz)

of a [0/22.5] square cantilever plate using lumped mass.
Grid size Mode 1 Mode 2 Mode 3 Mode 4 Mode 5

10x10 34.27 56.55 121.80 213.75 240.19
12x12 34.32 56.57 122.47 214.81 242.57
14x14 34.35 56.58 122.88 215.45 243.98
16x16 34.37 56.59 123.14 215.87 244.89
18x18 34.39 56.59 123.33 216.16 245.50
Kapania and
Singhvi (1991) 34.26 56.34 123.40 216.17 246.57

of a [0/22.5] square cantilever plate using consistent mass.

10x10 34.44 56.73 124.91 217.62 249.55
12x12 34.44 56.70 124.64 217.53 249.09
14x14 34.44 56.68 124.48 217.47 248.77
16x16 34.44 56.66 124.37 217.43 248.55
18x18 34.44 56.65 124.30 217.39 248.40
Kapania and
Singhvi (1991) 34.26 56.34 123.40 216.17 246.57
Table 3.14: Convergence of the first five natural

frequencies (Hz) of a [152 /0]s skew cantilever plate
(β = 30) using lumped mass.

4x8 16.81 74.68 105.82 193.29 284.36
6x12 16.87 75.98 108.65 197.22 294.52
8x16 16.89 76.45 109.70 198.65 298.26
10x20 16.90 76.67 110.20 199.32 300.02
12x24 16.91 76.79 110.47 199.68 300.98
Kapania and
Singhvi (1991) 16.96 77.28 111.40
Table 3.15: Convergence of the first five natural

frequencies (Hz) of a [152 /0]s skew cantilever plate
(β = 30) using consistent mass.

4x8 16.93 76.94 111.67 199.97 304.49
6x12 16.92 77.01 111.37 200.27 303.80
8x16 16.92 77.03 111.26 200.38 303.55
10x20 16.92 77.04 111.20 200.43 303.43
12x24 16.92 77.05 111.17 200.46 303.35
Kapania and
Singhvi (1991) 16.96 77.28 111.40

of a [0/22.5] skew cantilever plate (β = 45) using lumped mass.

10x10 20.58 44.89 115.10 142.33 216.67
12x12 20.57 45.00 115.47 143.25 218.02
14x14 20.56 45.06 115.69 143.80 218.83
16x16 20.55 45.10 115.82 144.14 219.35
18x18 20.55 45.13 115.91 144.38 219.71
Kapania and
Singhvi (1991) 20.42 45.09 115.75 144.76 219.94

of a [0/22.5] skew cantilever plate (β = 45) using consistent mass.

10x10 20.68 45.33 117.06 145.99 222.18
12x12 20.64 45.30 116.83 145.81 221.85
14x14 20.61 45.29 116.69 145.68 221.65
16x16 20.59 45.28 116.59 145.59 221.51
18x18 20.58 45.27 116.52 145.52 221.41
Kapania and
Singhvi (1991) 20.42 45.09 115.75 144.76 219.94
Table 3.18: Convergence of first five non-dimensionalized

frequencies (ωa2 ρ/E2 /h) of a simply supported [0/90/90/0]
spherical shell using lumped mass.

full shell
10x10 20.62 32.70 56.27 58.98 61.37
16x16 20.53 32.53 56.66 58.60 61.96
20x20 20.51 32.50 56.75 58.52 62.11
24x24 20.50 32.48 56.80 58.48 62.20
28x28 20.49 32.46 56.83 58.45 62.25
Reddy (1984) 20.38
Table 3.19: Convergence of first five non-dimensionalized

frequencies (ωa2 ρ/E2 /h) of a simply supported [0/90/90/0]
spherical shell using consistent mass.

full shell
10x10 20.58 32.65 56.31 58.94 61.42
16x16 20.48 32.48 56.67 58.55 61.95
20x20 20.46 32.44 56.75 58.47 62.09
24x24 20.45 32.42 56.80 58.43 62.17
28x28 20.44 32.41 56.83 58.41 62.22
Reddy (1984) 20.38
Table 3.20: Convergence of mid-deflection wmax due

to a nonuniform thermal field (Tu = 1, Tl = −1) and
maximum in-plane displacements umax , vmax due to a
uniform thermal field (Tu = 1, Tl = 1) for a simply
supported [0/90/90/0] square plate.
Grid size 105 umax 105 vmax 103 wmax

full plate
6x6 0.44274 0.44274 0.21177
8x8 0.44309 0.44309 0.21254
10x10 0.44329 0.44329 0.21286
12x12 0.44342 0.44342 0.21303
14x14 0.44350 0.44350 0.21312
16x16 0.44355 0.44355 0.21318
18x18 0.44359 0.44359 0.21322
20x20 0.44362 0.44362 0.21325
22x22 0.44364 0.44364 0.21327
Wu and
Tauchert (1980) 0.44138 0.44138 0.21336
Table 3.21: Convergence of mid-deflection wmax and

maximum in-plane displacements umax, vmax for a simply
supported [0/90/0/90] square plate (BC1) under a non-
uniform thermal field (Tu = 1, Tl = 0).

full plate
6x6 0.25268 0.19543 0.10995
8x8 0.25317 0.19544 0.11044
10x10 0.25341 0.19546 0.11065
12x12 0.25354 0.19548 0.11076
14x14 0.25363 0.19549 0.11083
16x16 0.25368 0.19550 0.11087
18x18 0.25372 0.19551 0.11090
20x20 0.25375 0.19552 0.11092
Wu and
Tauchert (1980) 0.25250 0.19446 0.11099

supported [15/-15/15/-15] square plate (BC2) under a
non-uniform thermal field (Tu = 1, Tl = 0)

full plate
6x6 0.16902 0.13556 0.09988
8x8 0.17326 0.13785 0.10025
10x10 0.17536 0.13860 0.10040
12x12 0.17660 0.13890 0.10047
14x14 0.17739 0.13903 0.10052
16x16 0.17795 0.13909 0.10055
18x18 0.17835 0.13911 0.10057
20x20 0.17864 0.13911 0.10058
Wu and
Tauchert (1980) 0.17967 0.13740 0.10063

supported [0/90/0/90] spherical shell (BC1) under a

full shell
6x6 0.15838 0.12010 0.79489
8x8 0.14889 0.11464 0.79315
10x10 0.15084 0.11832 0.79232
12x12 0.15219 0.11493 0.79186
14x14 0.14918 0.11697 0.79159
16x16 0.15077 0.11656 0.79141
18x18 0.15105 0.11609 0.79128
20x20 0.14944 0.11671 0.79119
22x22 0.15065 0.11572 0.79113

supported [45/-45/45/-45] spherical shell (BC1) under a
Grid Size 105 umax 105 vmax 104 wmax

full shell
6x6 0.14844 0.14844 0.71711
8x8 0.14208 0.14209 0.71545
10x10 0.14076 0.14077 0.71467
12x12 0.14322 0.14323 0.71423
14x14 0.14178 0.14178 0.71394
16x16 0.14131 0.14132 0.71374
18x18 0.14255 0.14256 0.71359
20x20 0.14196 0.14197 0.71348
22x22 0.14168 0.14169 0.71339
26x26 0.14196 0.14197 0.71326
30x30 0.14217 0.14218 0.71318
Z
C
mm
Sy
S y
m
X Y
m
D
B
L/2
40
o R A
R = 25
Boundary conditions:
L = 50
AB : Free
Thickness = 0.25
AD : v , w ,x = 0
E = 4.32 x 108
=0
Self weight : 90 per unit area
Figure 3.1: Geometry of the Scordelis-Lo roof.

F/4 Z
C Y X
m m
S y
D
Sy
mm
R = 300
L = 600
E = 3 x 106
R = 0.3
B F=1
Thickness = 3.0
L/2
A
AB : Symm
AD : v , w , x = 0
(rigid diaphram)
Figure 3.2: Geometry of the pinched cylinder

Z
o
18
D C
Y
X
sym
m
m
sym
-F/2
B
F/2
A
free
Radius = 10
7 DC : free or w = 0
E = 6.825 x 10
= 0.3
Figure 3.3: Geometry of the hemispherical shell


Qu
ar
ter
ch
or
dl
ine
Figure 3.4: Geometry of the skew cantilever plate.

Mode 2 (427.91 Hz) 0.49 Mode 3 (1026.11 Hz) 0.49

0.39 0.39
0.29 0.29
0.19 0.19
0.09 0.09
-0.01 -0.01
-0.11 -0.11
-0.21 -0.21
-0.31 -0.31
-0.41 -0.41
-0.50 -0.50
-0.60 -0.60
-0.70 -0.70
-0.80 -0.80
-0.90 -0.90
Mode 4 (1426.31 Hz) 0.49 Mode 5 (1731.61 Hz) 0.49

0.39 0.39
0.29 0.29
0.19 0.19
0.09 0.09
-0.01 -0.01
-0.11 -0.11
-0.21 -0.21
-0.31 -0.31
-0.41 -0.41
-0.50 -0.50
-0.60 -0.60
-0.70 -0.70
-0.80 -0.80
-0.90 -0.90
Figure 3.5: Normalized modes 2,3,4 and 5 of a [0/45/-45/90/90/-45/45/0] cantilever plate

obtained using consistent mass matrix
0.49 0.49
Mode 2 (56.65 Hz) Mode 3 (124.30 Hz)
0.39 0.39
0.29 0.29
0.19 0.19
0.09 0.09
-0.01 -0.01
-0.11 -0.11
-0.21 -0.21
-0.31 -0.31
-0.41 -0.41
-0.50 -0.50
-0.60 -0.60
-0.70 -0.70
-0.80 -0.80
-0.90 -0.90
0.49 0.49
Mode 4 (217.39 Hz) Mode 5 (248.40 Hz)
0.39 0.39
0.29 0.29
0.19 0.19
0.09 0.09
-0.01 -0.01
-0.11 -0.11
-0.21 -0.21
-0.31 -0.31
-0.41 -0.41
-0.50 -0.50
-0.60 -0.60
-0.70 -0.70
-0.80 -0.80
-0.90 -0.90
Figure 3.6: Normalized modes 2,3,4 and 5 of a [0/22.5] cantilever plate obtained using
consistent mass matrix
0.49 0.49
Mode 2 (77.05 Hz) Mode 3 (111.17 Hz)
0.39 0.39
0.29 0.29
0.19 0.19
0.09 0.09
-0.01 -0.01
-0.11 -0.11
-0.21 -0.21
-0.31 -0.31
-0.41 -0.41
-0.50 -0.50
-0.60 -0.60
-0.70 -0.70
-0.80 -0.80
-0.90 -0.90
0.49 0.49
Mode 4 (200.46) Mode 5 (303.35 Hz)
0.39 0.39
0.29 0.29
0.19 0.19
0.09 0.09
-0.01 -0.01
-0.11 -0.11
-0.21 -0.21
-0.31 -0.31
-0.41 -0.41
-0.50 -0.50
-0.60 -0.60
-0.70 -0.70
-0.80 -0.80
-0.90 -0.90
Figure 3.7: Normalized modes 2,3,4 and 5 of a [152 /0]s skew cantilever plate (β = 30)
0.49 0.49
Mode 2 (45.27 Hz) Mode 3 (116.52 Hz)
0.39 0.39
0.29 0.29
0.19 0.19
0.09 0.09
-0.01 -0.01
-0.11 -0.11
-0.21 -0.21
-0.31 -0.31
-0.41 -0.41
-0.50 -0.50
-0.60 -0.60
-0.70 -0.70
-0.80 -0.80
-0.90 -0.90
0.49 0.49
Mode 4 (145.52 Hz) Mode 5 (221.41 Hz)
0.39 0.39
0.29 0.29
0.19 0.19
0.09 0.09
-0.01 -0.01
-0.11 -0.11
-0.21 -0.21
-0.31 -0.31
-0.41 -0.41
-0.50 -0.50
-0.60 -0.60
-0.70 -0.70
-0.80 -0.80
-0.90 -0.90
Figure 3.8: Normalized modes 2,3,4 and 5 of a [0/22.5] skew cantilever plate (β = 45)
Mode 2 0.028 Mode 3 0.028

2 1/2 0.026 2 1/2 0.026
a (/E2) /h = 32.41 0.024
a (/E2) /h = 56.83 0.024
0.022 0.022
0.021 0.021
0.019 0.019
0.017 0.017
0.015 0.015
0.013 0.013
0.011 0.011
0.009 0.009
0.007 0.007
0.006 0.006
0.004 0.004
0.002 0.002
Mode 4 0.028 Mode 5 0.028

2 1/2 0.026 2 1/2 0.026
a (/E2) /h = 58.41 0.024
a (/E2) /h = 62.22 0.024
0.022 0.022
0.021 0.021
0.019 0.019
0.017 0.017
0.015 0.015
0.013 0.013
0.011 0.011
0.009 0.009
0.007 0.007
0.006 0.006
0.004 0.004
0.002 0.002
Figure 3.9: Normalized modes 2,3,4 and 5 of a [0/90/90/0] simply supported spherical shell
m C
Sym
D X Y
Sy
mm
B
L/2
R A
BC : Free
E1 = 3.3 Gpa AB : Hinged
E2 = 1.1 Gpa (u , v , w , y , z = 0)
12 = 0.25
G12 = 0.66 Gpa R = 2.54 m
L = .508 m
= 0.2 rad
Thickness = .0126 m
Figure 3.10: Geometry of the hinged cylindrical panel.

This study (16x16 mesh)

STRI3 (16x16 mesh)
1500 S4R (16x16 mesh)
1000
P (N)
500
0
0.000 0.010 0.020
Mid deflection (m)

Figure 3.11: Snap-through behavior of a [45/-45] hinged cylindrical panel under a
concentrated mid-load.
This study (N = 50 , i = 200)

STRI3 (N = 50 , i = 313)
6000
S4R (N = 50 , i = 245)
S4R5 (N = 50 , i = 212)
5000
N : No. of steps i : No. of iterations
Tip moment (N m)
4000 Initial shape and deformed

shape at tip moment = 2000
3000 Deformed
2000 Init
i al
1x20 mesh
1000
0
0 1 2 3 4 5 6 7
Tip deflection (m)
Figure 3.12: Cantilever beam under a tip moment.

This study (N = 32 , i = 120)

STRI3 (N = 90 , i = 283)
800 STRI3 (N = 37 , i = 201)
S4R (N = 18 , i = 87)
700 Brank et al (1996)

600
Sy
d
m
xe
m
Fi
500 P/2
P (N)
400
Sy
ee
m
Fr
m
300
200
Deformed shape
100
(P = 719.36 N , w/R = 0.9765)
0
0.0 0.2 0.4 0.6 0.8 1.0
w/R
Figure 3.13: Clamped cylindrical shell under pinching loads (quarter shell model, 30x30
mesh).
This study (N = 25 , i = 95)

S4R (N = 19 , i = 90)
S4R5 (N = 19 , i = 78)
60 Brank et al (1996)
(32x32, 4-node shell elements)
50 N : No. of steps i : No. of iterations

Sym P/4
m
P (N)
w
ee
40
Fr
m
m
Sy
30
Sym
m
20
Deformed shape
(P = 60.08 N , w = 2.818 m)
10
0
0 1 2
w (m)
Figure 3.14: Cylindrical shell with free ends under stretching loads (1/8th model, 20x20
mesh).
N = no. of steps , i = no. of iterations
400 This study (N = 22 , i = 88)

STRI3 (N = 48 , i = 267)
S4R (N = 25 , i = 131)
Buechter and Ramm (1992)
(16x16, 4-node elements)
300
D
C
2P (N)
Sy
m
Symm
m
200
B
P
A
e
P Fre
100 C-D:w=0
uA : In-plane displacemrnt at A
0
-8 -6 -4 -2 0
uA (m)
Figure 3.15: Hemispherical shell under pinching and stretching loads (u at point a, quarter
shell model, 40x40 mesh).

This study (N = 22 , i = 88)
400
STRI3 (N = 48 , i = 267)
S4R (N = 25 , i = 131)
wA : Transverse displacement at A
c-d:w=0
300
D
2P (N)
Sy
m
Symm
m
200 B
P
A
e
P Fre
100 Deformed shape

P = 202.5 N
wa = 0.0014 m
ua = 7.984 m
0
-0.8 -0.6 -0.4 -0.2 0.0
wA (m)
Figure 3.16: Hemispherical shell under pinching and stretching loads (w at point a, quarter
shell model, 40x40 mesh).
Line load (P N/m) along A-B

4
A B
3
P (N/m)

1 Deformed shape
P = 4.034 N/m This study (N = 44 , i = 223)
wb = 13.74 m STRI3 (N = 242, i = 1199)
S4R (N = 40 , i = 207)
Buechter and Ramm (1992)
(2x16 16-node elements)
0
0 5 10
wB (m)
Figure 3.17: Ring plate under line load at the free edge (full plate, 10x80 mesh).
[45/-45] This study

-3
Time step = 0.5x10 sec STRI3
Quarter shell model (16x16 mesh) S4R
0.040
0.030
Mid-deflection (m)
0.020
0.010
0.000
0.00 0.05 0.10
Time (secs)
Figure 3.18: Dynamic analysis of an anisotropic ([45/-45]) hinged cylindrical panel under a
concentrated load at the center.
o
Tcr = 39.53 C This study
o
Tcr = 38.60 C Meyers and Hyer
1.0 [45/-45/0/90]s
0.15x0.15x0.001016 m
wmax / h
B.C’s : u,v,w,z = 0
on all edges
0.5
w0 : max. magnitude of imperfection
This study (wo/h = .0005)

This study (wo/h = 0.05)
Meyers and Hyer (1992)
(Rayleigh-Ritz , wo/h = 0.05)
0.5 1.0 1.5 2.0 2.5 3.0

T / Tcr
Figure 3.19: Thermal postbuckling analysis of a simply supported square plate under uniform
temperature rise (full plate, 16x16 mesh).
600
Quarter shell model (20x20 mesh)
This study (N = 20 , i = 94)

500 STRI3 (N = 20 , i = 52)
S4R (N = 20 , i = 49)
Temperature (o C)
400
300
200
100
[22.5/-22.5/22.5/-22.5]
-3 -2 -1 0 1
wmax / h
Figure 3.20: Thermal postbuckling analysis of a clamped cylindrical panel under uniform
temperature rise.
20 Initial shape and deformed

3
shape at pL /EI = 20.148
deformed Z, w
Y
15
X ,u
d
xe
Fi i ni t i a
pL3/EI
l
1x20 mesh E = 1.2e10 N/m
2
10 = 0.0
L = 10 m
2
Area of C.S = 1x.1 m
5 N: no. of steps , i: no. of iterations

u: This study (N=31 , i=141)
w: This study (N=31 , i=141)
u: S4R (N=18 , i=142)
w: S4R (N=18 , i=142)
0
0.0 0.5 1.0
-u/L, w/L at tip
Figure 3.21: Cantilever beam under uniform external pressure.

20
This study
Case 1 (N = 31 , i = 141)
Case 2 (N = 93 , i = 452)
15
N : no. of steps , i : no. ot iterations
pL /EI
Case 1 : with kp
3
10
Case 2 : without kp
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

w/L at tip
Figure 3.22: Cantilever beam under uniform external pressure (comparison of results
obtained with and without pressure stiffness).
EI = 35000 kN cm2
This study
EA = 420000 kN
2.0 Seide (1974) R = 100 cm
2
Area of C.S = 1x0.1 cm
1.5
u = 0 at A
v = 0 everywhere
symm. conditions at
P0 R3/EI
B and C
1.0 wA : Radial deflection at apex A
Pressure distribution: P0 (1-cos(2))

3
Deformed shape at P0R /EI = 2.033
z,w
A
0.5 y,v x,u
R

B C
0.0
-0.5 -0.4 -0.3 -0.2 -0.1 0.0
wA/R
Figure 3.23: Thin circular ring under nonuniform pressure (half-ring model, 1x40 mesh, 80
elements).
Chapter 4
Shape Control of a Mirror Segment
4.1 Introduction and Literature Review

One of the most stringent requirements in the design of astronomical telescopes is the
maintenance of the primary mirror surface to a very high degree of accuracy. The shape of the
primary mirror has to be continuously monitored and any deviation from the desired shape
caused by external disturbances needs to be controlled by applying corrective measures.
Active control techniques are widely being adopted to correct the surface inaccuracies of
optical surfaces. This involves correction of the surface deformation by applying control
inputs using discrete or distributed actuation. In the case of discrete actuation, concentrated
forces are applied at discrete locations on the deformed mirror using actuators that are
located at the rear surface of the mirror. In the case of distributed actuation, strips of
piezoelectric material are used as actuators. These strips are either bonded to the rear
surface of the mirror or embedded in a composite mirror faceplate. Piezoelectric materials
undergo mechanical deformation when an electric field is applied. The resulting induced
distributed strain is used to control the surface deformation of the structure.
The active optics concept was first applied by Robertson et al. [91] to a 0.508 m diameter,
3-segmented mirror. The mirror segments were aligned using 3 actuators per segment, which
exerted control in piston and tilt only. Since the mirror segments were moved as rigid bodies
the elastic deformations could not be controlled using such control techniques. The concept
of thin deformable mirror was introduced by Robertson [92] who investigated analytically
and experimentally the feasibility of controlling the shape of a 0.782 m diameter 0.0127 m
100
CHAPTER 4. SHAPE CONTROL OF A MIRROR SEGMENT 101
thick spherical mirror using actuators that worked on the rear surface of the mirror. He used
58 equally spaced actuators to correct the surface error introduced during fabrication and
mounting the mirror on the reaction support system. A finite element model was utilized
to determine the actuator force configurations that would provide local displacements of the
mirror at the actuator control points. The initial rms figure error which was greater than
0.5 wavelength (6328 ×10−10 m) was corrected to within 0.02 wavelength.
Creedon and Lindgren [93] employed the modal expansion technique to represent the
surface distortion of a flexible structure in terms of its natural vibration modes and countered
the distortion by exerting control over a finite number of most significant modes. The
actuators were located at the nodes of the next highest mode. They demonstrated this
concept using flat rectangular and circular plates for which mode shapes are available in
closed forms. Howell and Creedon [94] extended the modal expansion method to a 0.762 m
diameter 0.0127 m thick spherical mirror used in the study by Robertson [92]. They used
the first 58 modes for the modal expansion and 58 equally spaced actuators as was done
by Robertson et al. [91]. The mode shapes were obtained using the finite element model
similar to that used by Robertson [92]. The actuator locations resulting in minimum surface
error were selected from the common nodes of a predetermined number of uncontrolled
modes. Ostroff [95] presented an improved method for selection of actuator locations based
on minimization of that part of the potential energy which is associated with the uncontrolled
modes. Ostroff used the first 48 modes for the modal expansion of the surface distortion.
The same finite element model which was used by Robertson [92] was employed but all the
available nodal points were considered for positioning the actuators.
Scott [96] proposed a new method in which the correction of the surface distortion of
the mirror was achieved by the application of moments at selected locations. An advantage
of this method is that equal and opposite moments could be applied, eliminating the need
for a rigid backup structure. Scott used a finite element model of a 0.254 m flat disc to
demonstrate the control of shapes in the form of the first 10 terms of the Zernike series,
employing a maximum of 24 moment actuators. Bushnell [97] studied the effectiveness of
concentrated loads in controlling the surface distortions of 4m diameter 0.01m thick spherical
caps and circular plates using a finite difference model. Two different distortions, one due to
a nonuniform thermal field given in terms of a few terms of the Zernike series and the other
consisting of Zernike polynomials themselves, were considered in his study. As many as 100
actuators were used to bring down the initial rms surface error by 2 orders of magnitude.
Bushnell found that mirrors with a very large diameter-to-thickness ratio and radius of
curvature-to-thickness ratio are difficult to control.
Hansen et al. [98] presented an integrated active mirror system with 41 actuators in which
a 4m diameter 0.02m thick spherical mirror was controlled in normal position, tangential
slope and radial slope. In their analytical study employing a finite element model, the
actuator system was evaluated to control deformations due to a thermal gradient through the
thickness, uniform thermal soak of the entire system, gravity loads and optical aberrations
like defocus and astigmatism. Hansen et al. [98] also found that thick mirrors are easier
to control than thin mirrors. They also came to interesting conclusions that deformations
due to bending loads like gravity or axial thermal gradient and deformations into shapes
of nondevelopable surfaces (defocus) are difficult to control, whereas deformations due to
membrane loads like a uniform rise in temperature and deformations into shapes of a
developable surface (astigmatism) are easy to control.
Masaki et al. [99] performed shape control experiments using a 0.62 m diameter 0.021 m
thick spherical mirror. They expressed the mirror deformations in the form of fundamental
Zernike modes (astigmatism, coma and spherical aberration) in terms of the first 24 normal
modes of the mirror and controlled only 7 or 12 out of the 24 observed modes using 9 force
control actuators. The initial rms figure errors of the order of a few hundred nanometers
could be reduced to about 18 nm when 7 modes were controlled. When 12 modes were
controlled the residual rms error was about 72 nm.
In the majority of the aforementioned works, the surface deformations were controlled
by applying forces at discrete locations on the rear surface of the mirror. A major drawback
of using such discrete devices is the requirement of a rigid and consequently heavy support
structure which is undesirable for space applications. An alternative to this approach is
to use piezoelectric actuators which can be very lightweight and do not require a heavy
backup structure. Kuo and Bruno [100] investigated the problem of determining the optimal
locations of piezoelectric actuators to control surface distortions of a one meter hexagonal flat
panel using a modified simulated annealing technique. They used two types of distortions,
one a linear combination of first the 3 normal modes and another a linear combination of first
the 5 normal modes, both having a maximum deformation of 5 microns in each mode. The
optimization was performed several times using different starting configurations to determine
the optimal configuration. They found that strips placed in the radial directions appeared in
almost every run. Hence a configuration with actuators placed along 6 radial lines resulting
in 24 actuators was used to control both types of distortions. Using such a configuration, the
rms error in the first case was reduced to 0.0305 microns and in the second case to 0.3529
microns. As can be seen from the results obtained by Kuo and Bruno [100], the configuration
of radially located piezoelectric actuators does not seem to be very effective in controlling
complex surface distortions of the mirror segment.
In a subsequent study, Kuo [101] used a similar configuration with 6 radial groups of 5
actuators in each group to excite the first 3 natural modes of vibration of the panel and
experimentally verified the numerical results obtained using the NASTRAN program. Using
the same configuration, Kuo [102] performed experiments on a 0.5 m diameter, 0.025 m thick
curved hexagonal mirror segment, demonstrating that by applying different voltages to the
six groups of actuators, the mirror could be deformed into shapes that are similar to the first
8 Zernike polynomials.
Though the numerical and experimental work done by Kuo and Bruno [100] and Kuo
[101],[102] demonstrate the use of piezoelectric actuators to control the surface distortions
of a mirror segment, the results presented do not give an indication of the effectiveness of
the configuration of radially located actuators in controlling complex surface distortions of
the mirror segment.
In the present work, control of thermal deformations of a 0.5 m hexagonal curved mirror,
0.012 m thick, with a radius of curvature of 10 m, using discrete and distributed actuators
is presented. The mirror segment used is part of a multisegmented primary mirror of a next
generation space telescope proposed by Jakubowski et al. [65]. A finite element model is used
to represent the mirror-actuator system. In order to study the effectiveness of the discrete
and distributed actuators, a comparative study is conducted using two different models of
the mirror-actuator system. In Model-1 the mirror is mounted on kinematic supports and
controlled by piezoelectric strips that are bonded to the rear surface of the mirror. Two sets
of strip configurations are studied. One with radially placed strips similar to that employed
by Kuo and Bruno [100] and the other with evenly distributed strips. In Model-2 the mirror
is mounted on force actuators which are used to support the mirror as well as control the
surface deformations of the mirror. The optimal values of the forces that are to be applied
at the actuator control points and the voltages to be applied across the piezoelectric strips
are determined by minimizing the rms surface figure error. A study is also conducted to
compare the performance of evenly distributed strips to that of strips placed at near optimal
locations obtained using heuristic integer programming. Details of the analysis and the
results are presented in the following sections.
4.2 Finite Element Model

The mirror is modeled using an assembly of flat triangular shell elements described in
Chapter 2. The piezoelectric strip is also modeled using the same flat triangular shell
element, but as a separate layer of isotropic material bonded to the lower surface of the
mirror. The force actuator is modeled using a 2-node beam element (Hughes [105]). The
beam element has bending rigidity in 2 orthogonal planes normal to its longitudinal axis,
extensional rigidity along its longitudinal axis and torsional rigidity about its longitudinal
axis giving a total of 6 degrees of freedom per node (3 translations and 3 rotations). Shear
deformations are also included in the beam formulation. One end of the actuator or the
beam element is constrained in all degrees of freedom to simulate the rigid backup structure
and the other end is connected to the mirror where the control force is applied. Thermal
deformations of the force actuators have not been considered in this study. Also the thermo-
elastic coupling in the mirror and the force actuators and the piezo-thermo-elastic coupling
in the piezoelectric strips have been neglected.
To demonstrate the correctness of the finite element formulation, an example of a
Graphite/Epoxy square plate and a piezoelectric bimorph cantilever beam are presented.
Several examples of plates and shells under mechanical and thermal loads have already been
presented in Chapter 3 and hence are not reproduced here.
Graphite/Epoxy laminate under an electric field: The plate is made of 8 layers ([0/90/0/90]s )
of graphite/epoxy material and a 9th layer of piezoelectric polymer material, polyvinylidene
fluoride (PVDF) twice as thick as the other 8 layers. The plate is simply supported (BC1 of
Chapter 3) on two opposite edges and clamped on the other two edges. A constant voltage
(Ez V/m) is applied across the PVDF layer. The properties of the graphite/epoxy laminate
(Jonnalagadda et al. [103]) are E1 = 181 GP a, E2 = 10.3 GP a, G12 = 7.17 GP a, ν12 = 0.28,
α1 = 0.02x10−6 /o C, α2 = 22.5x10−6 /o C and α12 = 0. Material properties for the PVDF
layer (Jonnalagadda et al. [103]) are E1 = E2 = 2 GP a, G12 = 0.752 GP a, ν12 = 0.33,
d31 = d32 = 23e−12 m/V and d33 = 33e−12 m/V , where d31 and d32 are the piezoelectric strain
coefficients which correspond to the in-plane strains that are induced due to a constant
electric field applied in the transverse direction. Note that the coefficient d33 corresponds
to the strain in the transverse direction and does not appear in the plane stress relations.
The author understands that it has been used by Jonnalagadda et al. [103] only to non-
dimensionalize the mid-deflection. The convergence of the non-dimensionalized center plate
deflection is given in Table 4.1 along with the results of Jonnalagadda et al [103] and are in
good agreement.
Piezoelectric bimorph cantilever beam: The beam (0.1 m x 0.005 m x .001 m) is made of two
layers of PVDF material with opposite polarity. A constant voltage of 1000 V/m is applied
across the thickness. The material properties are the same as those used in the previous
example, except that the Poisson ratio used in this case is zero. A 1x20 mesh (40 elements
and 42 nodes) was used to model the beam. The average of the transverse deflection at
the nodes along the two edges of the beam, at 5 locations along the beam, are presented
in Table 4.2 along with the analytical solution and finite element results (obtained using
10 solid isoparametric elements, 5 per layer) given by Tzou and Tseng [104]. The present
results agree well with the analytical solution given by Tzou and Tseng. The finite element
results of Tzou and Tseng based on an isoparametric solid element formulation are stiffer
compared to the analytical solution or the present results.
The mirror-actuator model was verified using an example of an L-frame fixed at the 2
ends and subject to concentrated forces Px , Py , Pz at point A (Fig 4.1) along the X, Y, Z
axes, respectively (Fig. 4.1), and a moment My at point A, about the Y-axis. The horizontal
member (member-1) of the L-frame was modeled using the flat shell elements with a Poisson’s
ratio of zero and the vertical member (member-2) was modeled using the beam element. The
results were compared with those obtained using Castigliano’s principle, those obtained by
modeling the horizontal member using the B33 2-node Euler-Bernoulli beam element and
the vertical member using the B31 2-node Timoshenko beam element of the commercial
finite element package ABAQUS and those obtained by modeling the horizontal member
using the 3-node flat shell element STRI3 of ABAQUS and the vertical member using the
B33 element. The STRI3 element of ABAQUS is a combination of the DKT plate bending
element and the Constant Strain Triangular (CST) element. The combination of STRI3
and B33 was used mainly to compare the in-plane deformations, as those obtained using
Castigliano’s principle or the beam elements (B33 and B31) are not expected to agree with
those obtained by modeling the horizontal member using the shell elements. The results are
presented in Table 4.3, where ux , uy and uz are the displacements along the X, Y and Z axes,
respectively. The present results are in good agreement with the rest of the results presented
in Table 4.3, demonstrating the validity of the present mirror-actuator model.
4.3 Control Algorithm

The surface deformations of the mirror segment are corrected by applying control inputs to
the force actuators and the piezoelectric strips which act upon the rear surface of the mirror.
In the present study, control input for the force actuator is the concentrated force that is
to be applied normal to the rear surface of the mirror at the grid points where the force
actuator is attached to the mirror. In the case of the piezoelectric strips, the control input
is the voltage to be applied across the thickness of the strips, which induces a distributed
strain in the strips and hence in the mirror, assuming that the strips are perfectly bonded
to the mirror. Since the number and locations of the force actuators and piezoelectric strips
are predetermined, the shape control problem involves the determination of the optimum
values of actuator forces or the voltages with which a maximum possible correction can be
obtained.
Consider a mirror approximated by m nodal points using the finite element model. Let
ψ denote the deformed shape or the transverse displacements and u the correction to the
transverse displacement applied using n force actuators or piezoelectric strips. In this study
the deformed shape ψ is assumed to be known. The deformed shape or the mirror figure
error can be estimated using the Shack-Hartmann wavefront analyzer [99]. For details the
user is directed to Ref. [99] and the references therein. It suffices to mention here that
the wavefront analyzer estimates the deformed surface of the mirror by measuring the light
emitted from the mirror. From the measurements made, the mirror surface is expressed in
terms of Zernike polynomials using a suitable curve fit and thus the mirror figure error at any
location is available as a linear combination of the Zernike polynomials. Though the Zernike
polynomials do not correlate to the mechanical characteristics of the mirror like the natural
mode shapes, they have been widely used in the existing literature to represent the shape of
the mirror. Masaki et al. [99] have showed that the lower Zernike modes are similar to the
natural modes of a spherical mirror and hence the mirror deformations can be expanded in
terms of the lower Zernike modes instead of the natural modes.
The correction ui at any nodal point is given by

n
ui = αij fj (4.1)
j=1
where fj is the force applied at the jth actuator location or the voltage applied across the
jth piezoelectric strip and αij is an influence coefficient defined as the deformation caused
at node i due to a unit force applied at the jth actuator alone or a unit voltage applied
across the jth piezoelectric strip alone. It is assumed that the material of the mirror, force
actuators and the piezoelectric strips is linear-elastic and isotropic. Since the deformations
considered in this study are of the order of a few micrometers, the geometric and material
nonlinearities are less significant and hence the superposition used in Eq. (4.1) is justified.
A matrix of influence coefficients of size mxn is obtained from the finite element model by
applying a unit load at each node where the actuator is attached to the mirror or by applying
a unit voltage across each of the piezoelectric strips, one at a time. The correction ui thus
obtained, when applied against the deformed structure, tends to nullify the deformations.
In order that the best possible correction is obtained, the deviation from the desired shape
should be made minimum at every point on the structure. Since the structure is modeled as
a set of finite number of nodal points, it is required that the deviation at these nodal points
be minimum. A measure of the overall deviation or the rms figure error is given by

1 m
E = (ψi + ui)2 (4.2)
m i=1
Substituting Eq. 4.1 in Eq. 4.2, the rms error can be expressed in terms of the unknowns
fj as follows:

1 m n 2

E = (ψi + αij fj ) (4.3)
m i=1 j=1
Setting ∂E/∂fk = 0 gives

m
n
(ψi + αij fj )αik = 0 k = 1, 2....n (4.4)
i=1 j=1

m
m
Eq. 4.4 is of the form [A]{f} = {b}, where Akj = αij αik , bk = − ψiαik . The solution
i=1 i=1
of this set of equations, obtained using the standard LDLT direct factorization, gives the
values of forces to be applied at n actuator locations or the voltages to be applied across

n piezoelectric strips for which the best possible correction can be obtained. Since the
material is assumed as linear-elastic, the control inputs thus obtained are those that cause
the undeformed mirror to deform to the shape given by ψ. The control inputs thus obtained
will have to be applied in the opposite direction, on the deformed mirror, in order to obtain
the necessary correction.
The control algorithm was validated by using the deformed shape due to arbitrary loads
applied at a few nodal points and that due to arbitrary voltages applied across a few
piezoelectric strips. The optimal values of the forces and the voltages obtained using the
control algorithm were exactly the same as those used to generate the deformed shape.
4.4 Results and Discussion

The aforementioned algorithm is applied to control the thermal deformations of a spherical
mirror segment with a hexagonal base. The geometry and material properties of the mirror,
piezoelectric strip (Ha et al. [106]) and force actuator are given in Table 4.4. The temperature
distribution at the lower surface of the mirror is assumed to be in the form of a linear
combination of the first few terms of the Zernike series (Pearson and Stepp, [107]) expressed
in terms of Cartesian coordinates x, y with the origin at the center of the mirror. The
Cartesian coordinates used to express the temperature distributions are normalized such
that they are in the range [-1,1]. The temperature distributions that are considered in
this study are given in Table 4.5, where the constant C is used to scale the temperature
distributions such that the upper, light-reflecting surface is at a lower temperature than the
lower surface with a constant temperature difference ΔTzo C, and the maximum temperature
o
difference between any two grid points across the lower surface of the mirror is ΔTxy C.
The finite element mesh for Model-1 (Fig. 4.2) consists of 864 flat shell elements and
469 grid points. The mirror in Model-1 is supported at the 6 vertices: 1, 13, 223, 247,
457 and 469. The translational degrees of freedom are constrained at the supports. It is
preferable to support the mirror on all six vertices than three vertices 120 degrees apart. This
is because the edge deformations in the former case resemble those of a simply supported
beam and are easier to correct than the cantilever type deformations along the edges in
the latter case. The thermal deformations are controlled using piezoelectric strips bonded
to the lower surface of the mirror. Two sets of configurations of the strips are considered,
one with 30 radially placed strips (Fig. 4.3) similar to the configuration used by Kuo and
Bruno [100] and the other with 121 evenly distributed strips (Fig. 4.3). It was observed that
better control could be achieved by adding strips near the periphery to a set of 79 evenly
distributed strips (Fig. 4.3) resulting in a set of 121 evenly distributed strips as shown in Fig.
4.3. Since the piezoelectric strips are very light and flexible, a large number of such strips
can be used without drastically increasing the overall weight or stiffness of the mirror unlike
the conventional force control actuators. The configuration with 121 strips as shown in Fig.
4.3 consists of 37 hexagonal strips and 84 semi-hexagonal strips. Each hexagonal strip has
a side of 0.04166 m. The optimal locations are determined from a starting configuration of
193 strips (Fig. 4.3) consisting of 91 hexagonal strips and 102 semi-hexagonal strips. The
heuristic algorithm used to determine the (near) optimal locations is described later in this
section.
In the case of radially placed and evenly distributed strips, the influence coefficients are
determined by applying a unit voltage across each strip, one at a time, in the presence of all
other strips. In order to determine the optimal locations of the strips, an approach similar to
that followed by Kuo and Bruno [100] is used. The influence coefficients are first determined
by neglecting the stiffness of the strips in computing the global stiffness matrix. If the
stiffness of the strips is included in the global stiffness matrix for determining the influence
coefficients, each set of influence coefficients would correspond to a different global stiffness
matrix and hence the linear combination of the sets of influence coefficients thus obtained
will not correspond to any one stiffness matrix. In other words, the influence coefficients are
determined by applying a unit voltage across each of the strips, one at a time, in the absence
of the rest of the strips.
The optimal locations of the strips are obtained using the Dlorenzo algorithm given by
Haftka and Adelman [108]. This algorithm has been chosen as it does not require any
initial guess of the locations as in other optimization techniques like simulated annealing
[100]. More exhaustive integer programming methods are available in the literature (Haftka
and Adelman [108]), but they are computationally much more expensive than the Dlorenzo
algorithm and they also have a certain degree of uncertainty due to the requirement of initial
guess locations.
The Dlorenzo algorithm starts with an initial configuration of 193 strips as shown in
Fig. 4.3. The steps followed in the algorithm are as follows: 1. Remove each one of the
strips from the initial configuration of n strips (n=193), one at a time, and determine the
rms error due to n configurations of n − 1 strips. 2. Rank the strips in ascending order of
the rms error. 3. Remove the strip which results in the least rms error, to obtain a new
configuration of n − 1 strips. 4. Repeat steps 1-3 until the desired number (30 or 121) of
strips is reached. The configuration thus obtained is called the optimal configuration in the
subsequent sections though it is certainly not the best possible configuration.
The thermal deformations of Model-1 are computed by neglecting the stiffness of the
strips. The rms and absolute values of the transverse deformation of Model-1 without the
o o
strips and with 30 and 121 strips are given in Table 4.6 for ΔTz =0.2 C and ΔTxy=0.5 C.
As can be seen from Table 4.6, there is not much difference in the rms and absolute maximum
deformations of the mirror without the strips and with 30 and 121 strips. This can be
attributed mainly to the fact that the thickness and the stiffness of the strips are much
lower than those of the mirror. Hence neglecting the stiffness of the strips in determining
the optimal locations is not expected to give erroneous results. Moreover, as the deformed
shape of the mirror without the strips did not vary significantly from that in the presence
of the strips, the optimal locations obtained as mentioned above could be used to control
the deformed shape in the presence of the strips. After determining the optimal locations
as mentioned above, the influence coefficients are computed now by including the stiffness
of the strips as done in the case of radial and evenly distributed strips.
The corrected rms figure error and the maximum residual deformations for Model-1 using
o o
30 radially placed strips and 30 optimally placed strips for ΔTz =0.2 C and ΔTxy =0.5 C
are given in Table 4.7. The optimal voltages to be applied to 30 strips varied from -558
V to 389 V in the case of radially placed strips, and in the case of optimally placed strips
the voltages varied from -1312 V to 853 V. The negative sign indicates that the voltage is
applied in the direction opposite to the direction of polarization of the piezoelectric material.
The corrected rms figure error and the maximum residual deformations for Model-1 using
o o
121 evenly placed strips and 121 optimally placed strips for ΔTz =0.2 C and ΔTxy =0.5 C
are given in Table 4.8. The optimal voltages to be applied to 121 strips varied from -983 V
to 547 V in the case of evenly placed strips, and in the case of optimally placed strips the
voltages varied from -1159 V to 1242V. The optimal locations of 30 and 121 strips are shown
in Figures 4.4 and 4.5, respectively.
The surface distortions could not be reduced much using 30 radially placed strips. The
poor performance of 30 radially placed strips can be attributed to the fact that the influence
of the strips is restricted to a certain area, unlike the case of evenly distributed strips. Even
using 30 (near) optimally placed strips, the percentage reduction in the rms figure error or
the maximum residual deformation seems insufficient in the case of complex deformations
T2 and T3 (Table 4.7). The performance of 121 evenly distributed strips (Table 4.8 ) is
quite close to that of 121 (near) optimally placed strips. The amount of correction obtained
in the case of asymmetric thermal fields T2, T3 and T4 indicate that the performance of 121
evenly distributed strips is not biased towards symmetric deformations or symmetric support
conditions. The corrected rms values given in Table 4.8 for 121 evenly distributed strips
range from 81 nm to 97 nm. The rms figure error of the surface distortions of the mirror
segment of the next generation space telescope (Jakubowski [65]) should be within 1/20th
of the operating wavelength (0.5 μm) in order to achieve satisfactory optical performance.
o o
By restricting ΔTz to 0.025 C and ΔTxy to 0.4 C, the initial rms figure errors ranging
from about 0.5 to 2.1 μm could be reduced to within 29 nm, using 121 evenly distributed
strips (Table 4.9). The optimum voltages applied to 121 evenly distributed strips in this case
range from -290 V to 238 V (-1.16 MV/m to 0.95 MV/m).
The finite element mesh for Model-2 consists of 864 flat shell elements and 61 beam
elements. The mirror is mounted on 61 force actuators (Fig. 4.6) which are used to support
the mirror as well as to control the surface deformations of the mirror. The actuators
are placed with an equal inter-actuator distance of 0.125 m. One end of the actuators or
beam elements is constrained in all 6 degrees of freedom and the other end is attached to
the mirror. The Young’s modulus of the actuators is assumed to be 1/100 of that of the
mirror. The uncorrected and corrected rms and maximum absolute values of the transverse
o
deformations of Model-2, in μm, due to different temperature distributions for ΔTz =0.2 C
o
and ΔTxy =0.5 C are given in Table 4.10. The optimum actuator forces applied to actuators
in this case varied from -34 N to 20.4 N. The uncorrected and corrected rms values for
o
ΔTz =0.025 C and ΔTxy =0.4o C, are given in Table 4.11. The optimum actuator forces
applied to actuators in this case varied from -6.4 N to 9 N.
Since the actuators are used for both support and control, changing the location of
the actuators will result in different support configurations and hence different thermal
deformations. In other words, the problem does not remain the same by changing the
locations of the actuators. So no attempt is made to determine the optimal locations.

The thermal deformations in the case of Model-2 are smaller than those of Model-1 due
to the obvious reason that the mirror is more constrained in Model-2 than in Model-1.
Except for T1, the corrected rms figure errors obtained using 61 force actuators are quite
close to those obtained using 121 piezoelectric actuators for ΔTz =0.2o C and ΔTxy =0.5o C.
The thermal deformation due to the temperature distribution T1 resembles that of defocus,
which can be corrected using a focal plane correction system (Masaki [99]) without using
o o
force actuators or piezoelectric strips. When ΔTz is reduced to 0.025 C and ΔTxy to 0.4
C, the percentage reduction in the rms error obtained using 61 force actuators is quite close
to that obtained using 121 piezoelectric strips, for all four temperature distributions. The
corrected rms in the case of the force actuators (Table 4.11) is much lower as the uncorrected
rms is also much lower than those in the case of the piezoelectric strips.
4.5 Summary
Both the force actuators and the piezoelectric strips were equally effective in controlling the
surface deformations of the mirror. The performance of 121 evenly distributed strips was
quite comparable to that of 121 optimally placed strips. Mounting the mirror on a large
number of force actuators (Model-2) appeared to be more advantageous than supporting the
mirror on six kinematic supports (Model-1) at the vertices, since the mirror deformations in
the former case were much lower than those in the latter, but a disadvantage of using the
force actuators is the increase in the overall weight of the structure, which is undesirable
for space applications. Since the piezoelectric strips are lightweight, a large number of such
strips can be used to control the surface distortions of the mirror without imposing a weight
penalty. Piezoelectric strips can also be used to correct both developable and non-developable
surfaces since they, unlike the force actuators, can be used to impart both membrane and
bending strains on the mirror.
In the actual mirror, the piezoelectric strips cannot be relocated as they will be bonded
to the rear surface of the mirror. Moreover, the optimal locations determined for one type of
disturbance may not be optimal for other disturbances. Correcting the mirror deformations
using a set of strips with predetermined locations thus seems to be more practical than trying
to find the optimal locations. Controlling the mirror deformations using a large number of
evenly placed piezoelectric strips seems to be a viable solution, as such a configuration can
impart control to almost every part of the mirror and hence can be used to control arbitrary
deformations.
Since the piezoelectric strips act over a finite area of the mirror and the force actuators
only act discretely, the mirror surface corrected using the former is expected to be smoother
than that corrected using the latter. The quality of the mirror surface should be assessed in
order to determine if the mirror surface is smooth enough to obtain the required operational
efficiency. The rms figure error may not be a true measure of surface smoothness. Hence
alternative measures to assess the surface smoothness of the mirror have to be determined.
It may not be possible to generate in space the kind of voltages, as determined in this
study, that are to be applied to the piezoelectric strips. Hence there is a need for piezoelectric
materials with much higher strain constants. A parametric study can also be conducted to
determine the optimum values of the stiffness and the dimensions of the mirror and the
piezoelectric strips which will result in the least thermal deformations and voltages to be
applied across the strips. The feasibility of using a large number of piezoelectric strips on
a mirror segment should also be examined, since the use of a large number of strips could
result in a very complex electronic circuitry. In summary, the piezoelectric strips appear
to be promising candidates for static shape control of flexible structures in space, opening
several avenues for further research in order to determine the feasibility of using a large
number of such strips for practical problems.
Table 4.1: Convergence of non-dimensionalized mid-

deflection of a graphite/epoxy laminate excited by a
uniform electric field.
Grid Size 100W (h/b)/(bd33 Ez )

quarter plate
2x2 0.14042
4x4 0.13996
6x6 0.13962
8x8 0.13945
10x10 0.13936
12x12 0.13933
14x14 0.13927
16x16 0.13925
18x18 0.13924
Jonnalagadda et al(1994)
4x4 mesh (quarter plate) 0.1396
9-node Lagrangian elements
Table 4.2: Transverse deflection (x107 ) in m of a piezoelectric bimorph

cantilever beam under uniform electric field, at different locations.
Location 2 cm 4 cm 6 cm 8 cm 10 cm
This study (1x20 mesh) .1376 .5516 1.2416 2.2076 3.4496
Tzou and Tseng (1990) FEM .124 .508 1.16 2.10 3.30
Tzou and Tseng (1990) Theory .138 .552 1.24 2.21 3.45
Table 4.3: Deflection in μm, at point A, of the L-frame
Px = 100N Py = 100N Pz = 100N My = 100N

Present ux = .291 ux = 0 ux = 0 ux = 0.0257
uy = 0 uy = 37.63 uy = 0 uy = 0
uz = 0 uz = 0 uz = 33.8 uz = −50.6
Theory ux = .284 ux = 0
uy = 0 uy = 0
uz = 0 uz = 33.8
B33+B31 ux = .284 ux = 0
uy = 0 uy = 0
uz = 0 uz = 33.8
STRI3+B33 ux = .291 ux = 0 ux = 0 ux = 0.0267
uy = 0 uy = 37.59 uy = 0 uy = 0
uz = 0 uz = 0 uz = 33.8 uz = −50.6
Table 4.4: Properties and geometry of the mirror, piezoelectric strip and
force actuator.
Property/geometry Mirror Piezoelectric Force actuator

(Beryllium) strip
Young’s modulus 293 GP a 63 GP a 2.93 GP a
Poisson’s Ratio 0.1 0.3 0.1
0 0
Coefficient of 11.5E-6/ C 0.9E-6/ C
thermal expansion
Thickness 0.012 m 0.25E-3 m
Radius of curvature 10 m
Side of the Hexagonal Base 0.5 m
d31, d32 254E-12 m/V
Diameter 0.008 m
Length 0.05 m
Table 4.5: Temperature distribution at the lower

surface of the mirror.
T1 C[2(x2 + y 2) − 1]
T2 C[(x + y)(3x2 + 3y 2 − 2)]
i=9
T3 C[ i=1 Ki Zi ]
Ki s from Table 2 (back surface) of Ref. 21
T4 C[x + y + 2xy]
Table 4.6: Uncorrected rms and maximum absolute

values of the transverse deformations of Model-1, in
μm, due to different temperature distributions (ΔTz =
0.2o C, ΔTxy = 0.5o C).
T(x,y) without strips with 30 strips with 121 strips

rms abs. rms abs. rms abs.
T1 4.85 9.65 5.01 9.89 5.00 9.88
T2 3.00 5.27 3.14 5.49 3.11 5.47
T3 1.89 4.91 1.89 4.87 1.87 4.81
T4 3.28 5.68 3.39 5.89 3.37 5.86
Table 4.7: Corrected rms and maximum absolute residual values of the
transverse deformations of Model-1, in μm, due to different temperature
distributions, using 30 piezoelectric strips (ΔTz = 0.2o C, ΔTxy = 0.5o C).
T(x,y) 30 radially placed strips 30 optimally placed strips

rms abs. rms abs.
T1 0.556 (11.5 %) 1.868 (19.3 %) 0.201 ( 4.1 %) 0.686 (7.1 %)
T2 0.749 (24.9 %) 2.698 (51.2 %) 0.220 ( 7.3 %) 0.884 (16.8 %)
T3 0.795 (42.0 %) 3.822 (77.8 %) 0.199 (10.5 %) 0.883 (17.9 %)
T4 0.924 (28.1 %) 3.578 (62.9 %) 0.217 ( 6.6 %) 0.840 (14.9 %)
Table 4.8: Corrected rms and maximum absolute residual values of

the transverse deformations of of Model-1, in μm, due to different
temperature distributions, using 121 piezoelectric strips (ΔTz =
0.2o C, ΔTxy = 0.5o C).
T(x,y) 121 evenly distributed strips 121 optimally placed strips

rms abs. rms abs.
T1 0.081 (1.7 %) 0.256 (2.7 %) 0.058 (1.2 %) 0.197 (2.0 %)
T2 0.093 (3.1 %) 0.255 (4.9 %) 0.075 (2.5 %) 0.248 (4.7 %)
T3 0.088 (4.6 %) 0.307 (6.3 %) 0.070 (3.7 %) 0.246 (5.0 %)
T4 0.098 (3.0 %) 0.309 (5.4 %) 0.079 (2.4 %) 0.223 (3.9 %)
Table 4.9: Uncorrected and corrected rms values of

the transverse deformations of Model-1, in μm, due
to different temperature distributions, using 121 evenly
distributed strips for ΔTz =0.025o C and ΔTxy =0.4o C.
T(x,y) Uncorrected rms Corrected rms

T1 1.971 0.0239 (1.2 %)
T2 0.536 0.0147 (2.7 %)
T3 2.125 0.0278 (1.3 %)
T4 1.195 0.0285 (2.4 %)
Table 4.10: Uncorrected and corrected rms and maximum absolute

values of the transverse deformations of Model-2, in μm, due to
different temperature distributions, using 61 force actuators (ΔTz =
0.2o C, ΔTxy = 0.5o C).
T(x,y) Uncorrected Corrected

rms abs. rms abs.
T1 0.5228 1.1997 0.0444 (8.5 %) 0.0954 (7.9 %)
T2 0.9456 2.4391 0.0414 (4.4 %) 0.0946 (3.8 %)
T3 1.1800 2.8037 0.0407 (3.4 %) 0.0942 (3.4 %)
T4 0.9212 2.1236 0.0413 (4.5 %) 0.0915 (4.3 %)
Table 4.11: Uncorrected and corrected rms values of

the transverse deformations of Model-2, in μm, due
to different temperature distributions, using 61 force
actuators (ΔTz = 0.025o C, ΔTxy = 0.4o C).
T(x,y) Uncorrected rms Corrected rms

T1 0.5119 0.0078 (1.5 %)
T2 0.2062 0.0054 (2.6 %)
T3 0.3802 0.0050 (1.3 %)
T4 0.1153 0.0052 (4.5 %)
Y
ed
Fix Member-1
X
Member-1 A
(flat shell elements) Member-2 Member-2
(beam element)
Length = 1 m Fixed
Width = 0.1 m Length = 0.05 m
Thickness = 0.012 m Radius = 0.004 m
E = 293 Gpa E = 2.93 Gpa
= 0.0 = 0.0
Figure 4.1: Geometry and material properties of the L-frame

457 469
443 456
428 442
412 427
395 411
377 394
358 376
338 357
317 337
295 316
272 294
248 271
223 247
199 222
176 198
154 175
133 153
113 132
94 112
76 93
59 75
43 58
28 42
14 27
1 13
Figure 4.2: Finite element mesh for Model-1.

Fig. 2a Fig. 2b
30 strips 79 strips
Fig. 2c Fig. 2d
121 strips 193 strips
Figure 4.3: Piezoelectric strip configurations.

T1 T3
T2 T4
Figure 4.4: Optimal locations of 30 strips.

T1 T3
T2 T4
Figure 4.5: Optimal locations of 121 strips.

514 518 522 526 530
463 467 471 475 479 483
402 406 410 414 418 422 426
331 335 339 343 347 351 355 359
250 254 258 262 266 270 274 278 282
173 177 181 185 189 193 197 201
106 110 114 118 122 126 130
49 53 57 61 65 69
2 6 10 14 18
Figure 4.6: Locations of 61 force actuators.

Chapter 5
Analysis of Inflatable Structures
The second application for which the flat shell element described in Chapter 2 is intended to
be used is the analysis of a large inflatable structure as shown in Fig. 1.1. Such a structure
is being considered by the US Army to be used as a transportable tent to house vehicles and
personnel. The updated Lagrangian formulation of the flat shell element described in Chapter
2 has been primarily developed to analyze large inflatable structures, as such structures are
expected to undergo large deformations under the action of environmental loads like wind
and snow loads. The tent structure is made up of arches which are stiffened by inflation, and
membranes which are connected to the arches and held under tension. Before attempting to
analyze the tent structure, a single arch under mechanical and pressure loads was analyzed,
in order to verify the accuracy of the present flat shell formulation in predicting the nonlinear
response of such large flexible structures. The results of this analysis are presented below.
5.1 Analysis of a Thin Circular Arch

Thin circular arch under a concentrated apex load: The geometry and the material
properties used are given in Fig. 5.1. One quadrant of the arch under a concentrated
apex load was analyzed. In Fig. 5.1, the results of the convergence study, performed using
the present element, are presented. As can be seen from Fig. 5.1, the present results are
pretty much converged for mesh sizes 25x80 and beyond. The results obtained using the
present element, STRI3, S4R5 and S4R, using a 25x100 mesh, are presented in Fig. 5.2.
As can be seen from Fig. 5.2, both STRI3 and S4R5 fail to give a stable solution, even
125
CHAPTER 5. ANALYSIS OF INFLATABLE STRUCTURES 126
using a fine mesh (25x100). The present element or S4R did not show such behavior and
the agreement between the two is quite comparable. As can be seen from Fig. 5.2, the arch
undergoes large deformation only in the region around the apex. Hence a finer grid could
have been used only around the apex. Such grid stretching was not done, as the main aim
of the analysis was to assess the accuracy of the formulation.
Thin circular arch under nonuniform pressure: The geometry and the material
properties used are given in Fig. 5.3. Two types of pressure distributions were considered, one
symmetric (Fig. 5.3) and another unsymmetric (Fig. 5.5). The same geometry and material
properties were used for both types of pressure distributions. The pressure load is assumed
to be space attached, or in other words the magnitude of the pressure distribution along
the length of the arch is treated as a function of the current/deformed configuration of the
arch. The magnitude around any cross section is assumed to be uniform. The analysis was
performed with the pressure stiffness matrix included in the system tangent stiffness matrix,
since it is clear from the example of the cantilever beam presented in Chapter 3 that the rate
of convergence of the solution is greatly enhanced if the pressure stiffness matrix is included.
In Fig. 5.3, the results of the convergence study for the symmetric pressure distribution
are shown. The convergence is quite rapid, unlike the case of the concentrated load. The
present results obtained using a fairly coarse mesh (20x40) appear to be a reasonably good
estimate of the converged solution, even in the regime of large rotations. The present results,
obtained using the arc-length control method, are compared with those obtained using the
S4R and STRI3 elements using the same solution procedure and convergence criterion, in
Figs. 5.4 and 5.5 for the two types of pressure distributions.
In the case of the symmetric pressure distribution, the analysis was performed until the
two diametrically opposite points of the cross section at the apex met each other. In the
case of the unsymmetric pressure distribution, the analysis was terminated at point B (Fig.
5.5) since the convergence was too slow even in the case of the curved shell element (S4R)
probably due to severe nonlinearities and large deformations.
As can be seen from Figs. 5.4 and 5.5, the present results are in good agreement with
those of S4R. The results obtained using STRI3 do not agree well with the present results
and those obtained using S4R, especially in the case of the unsymmetric pressure distribution
(Fig. 5.3). Poor performance of the STRI3 element can also be noticed in the case of the thin
circular arch under concentrated apex load (Fig. 5.1) and in the case of nonlinear response
of some shell problems presented in Chapter 3. The present flat shell formulation, which is
based on the assumption that the incremental rotations are moderate, seems more reliable
and requires fewer load steps compared to STRI3 which is based on a large rotation theory
similar to that proposed by Simo and Fox [53]. It is not clear why a large number of load
steps are required in the case of STRI3 in spite of the fact that the formulation is based
on the large rotation theory. The reason for the better performance of the present flat shell
element could be attributed to the use of a better membrane representation than that used
in the case of STRI3.
After having validated the present flat shell element for a wide variety of problems and
gaining sufficient confidence in using the element to obtain the nonlinear response of large
flexible structures, the element is currently being used for the nonlinear analysis of the tent
structure, under wind and snow loads. The wind load is being modeled as a non-uniform
pressure load and the snow load as lumped concentrated loads. Preliminary results of the
analysis of the tent structure are presented below.
5.2 Analysis of a Tent Structure

A symmetric model of the tent structure which represents a repeated combination of arches
and membranes was first analyzed in order to gain insight into the behavior of the structure
under the action of wind and snow loads. The material properties used are the same as those
used in the previous examples involving a single arch.
Shown in Fig. 5.6 are the results of the analysis of the tent structure under a wind load,
with a distribution as shown. One half of the arch and the membrane with symmertric
boundary conditions was analysed using a 20x20x120 mesh, with 20 divisions along one half
of the cross section of the arch (AF in Fig. 5.6), 20 along the width of the membrane (AB in
Fig. 5.6) and 120 divisions along the length of the arch/membrane (AD or BC in Fig. 5.6).
The point of attachment of the membrane to the arch is shown in Fig. 5.6. The arch was
first inflated to 100 psi and then the wind load was applied. The wind load was applied to
the entire membrane and portion of the arch not covered by the membrane (AG in Fig. 5.6).
The wind load was modeled as a nonuniform external pressure, with the direction normal
to the deformed congifuration. As expected, the tent undergoes large deformations near the
area where the wind load is like external pressure and also sways to the right due to the
suction on the rest of the tent.

Shown in Fig. 5.7 are the results of the analysis of the tent structure under a snow
load. A quarter model of the tent was analyzed using a 20x18x60 mesh. The geometry and
the material properties used are the same as those used in the case of the wind load. The
snow load was assumed to act over a sector 22.5o on either side from the apex, on the area
projected onto the XY-plane. As in the case of the wind load, the arch was inflated to 100
psi before applying the snow load. The apex deflection due to the initial inflation in the
case of the half and quarter tent models are slightly different, probably due to the symmetry
conditions along CDE in Fig. 5.7, which were not enforced in the half tent model (Fig. 5.6).
quarter arch model

2500
20x40 mesh
20x60 mesh
25x80 mesh
2000 25x100 mesh
40x100 mesh
50x100 mesh
Apex load P (lb)
1500
P/4
1000
r
A
R = 108 in
500 r = 6 in
R E = 1e6 lb/in2
= 0.3
h = 0.15 in
(thickness)
0 Fixed
-6 -5 -4 -3 -2 -1 0
Apex deflection (in), + ve upward
Figure 5.1: Convergence of the deflection at point A, of a thin circular arch under a
concentrated apex load P
2500
N : no. of steps , i : no. of iterations
This study (N = 44 , i = 180)

STRI3 (N = 81 , i = 401)
2000 S4R (N = 20 , i = 95)
S4R5 (N = 35 , i = 112)
Apex load P (lb)
1500
1000
500
Deformed shape of full arch

0 at P = 2465 lb
-7 -6 -5 -4 -3 -2 -1 0
Apex deflection (in), +ve upward
Figure 5.2: Deflection at point A, of a thin circular arch under concentrated apex load P
quarter arch model

6
20x40 mesh
20x60 mesh
25x65 mesh
25x80 mesh
P0 (psi)
r
A Pressure distribution:
P0 (1-cos2)
R = 102 in
2 r = 6 in
R E = 1e6 lb/in2
= 0.3
h = 0.15 in
(thickness)

Fixed
0
-5 -4 -3 -2 -1 0
Deflection at A (in), +ve upward
Figure 5.3: Convergence of the deflection at point A, of a thin circular arch under a symmetric
nonuniform external pressure
N : No. of steps , i : No. of iterations

This study ( N = 61 , i = 233 )
STRI3 ( N = 99 , i = 509)
8 S4R ( N = 38 , i = 181 )
6
Deformed shape of half arch
at P0 = 8.267 psi
5
P0 (psi)
3
w
2
1
Fixed
0
0 1 2 3 4 5
W (in)
Figure 5.4: Deflection at point A, of a thin circular arch under symmetric non-uniform
external pressure (quarter arch model, 25x65 mesh, 3250 elements, 10296 d.o.f.)
B
8
This study ( N = 28 )
STRI3 ( N = 90 )
S4R ( N = 20 )
6
p0 (psi)
N = no. of load steps

Analysis abandoned beyond point B
due to very slow rate of convergence
4 A
R
z,w
2

Fixed x Fixed
Deformed shape at p0 = 8.153 psi Pressure distribution: -p0 cos()
0
0.0 0.1 0.2 0.3 0.4
Deflection at the apex wA (in)
Figure 5.5: Deflection at point A, of a thin circular arch under unsymmetric nonuniform
pressure (half arch model, 25x150 mesh, 7500 elements, 23556 dof)
FE mesh (quarter model: ABCDEF) Point of attachment

Elements: 9600
of membrane to arch Wind Load Distribution
0.6 rad
Nodes : 4961
G
D.o.f. : 29766 A F
B.C’s :
ADEFA: symm
FAB: fixed
EDC: fixed
n C
BC: symm 2i D
10 E
cos(+0.6pi/2)
P() = P0
8 in
22 2
1+sin 1.2
96 B
in F
Thickness: 0.15 in
A
Load-Deflection Curve Deformed Shape
3
P0 (psi)
Mid-deflection due to initial inflation

of the arches by 100 psi : 0.0687 in Initial inflation (100 psi) +
1
wind load (P0 = 3.664 psi)
umax = 8.36 in
vmax = 0.45 in
0.06 0.08 0.10 0.12 0.14 wmax = 3.27 in
Mid-deflection (in)
Figure 5.6: Analysis of a tent structure under a wind load

FE mesh (quarter model: ABCDEF)

Elements: 4560
Nodes : 2379
D.o.f. : 14274
C
B.C’s : D
AFEDA: symm E
FAB: fixed 45o
EDC: symm
BC: symm n
2i
10
n Snow load:
8i
96 22 lb per projected area over a sector
in B 0
A F Thickness: 0.15 in 22.5 on either side from apex
Load-Deflection Curve Deformed Shape Z, w
0.8 X
0.7 Y
Snow load (lb/in2)
0.6
0.5
0.4
0.3
Apex deflection
0.2 due to inflation:
0.0722 in Inflation of arches (100 psi)
0.1
2
+ snow load (0.8226 lb/in )
-1.0 -0.5 0.0 0.5
Apex deflection (in) wmax = -5.8 in
Figure 5.7: Analysis of a tent structure under a snow load

Conclusions
Finite element analysis of general shells using a three-noded flat triangular shell element
was presented. The flat shell element is a combination of the Discrete Kirchhoff Theory
(DKT) plate bending element and a membrane element similar to the Allman element, but
derived from the Linear Strain Triangular (LST) element. The DKT element has three
degrees of freedom per node, a transverse displacement and two out-of-plane rotations. The
Allman element also has three degrees of freedom per node, two in-plane displacements and
an in-plane rotation. Though this combination has been employed in the literature for linear
static analysis of laminated plates, the results presented are not adequate to ascertain that
the element would perform well in the case of static and dynamic analysis of general shells.
In the present research, the element was first thoroughly tested for linear static analysis of
laminated plates and shells and , after having validated its accuracy, the element has been
extended for free vibration, thermal, and geometrically nonlinear analysis.
The major drawback of the DKT plate bending element is that the transverse
displacement is not explicitly defined within the interior of the element. Hence obtaining the
consistent mass matrix or the derivatives of the trasnverse displacement that are required for
forming the geometric stiffness matrix is not straightforward. This problem can alleviated
by using a lumped mass matrix, but the analysis is expensive, especially if a large number
of modes are to be extracted, due to slow rate of convergence. In the present research,
free vibration analysis was performed both by using lumped and so-called consistent mass
matrices, in order to compare the performace of the two methods. A fourth-order polynomial,
borrowed from an existing element, was used to compute the mass terms corresponding to
the transverse displacement. The same shape functions that were used to compute the
stiffness matrix of the LST element were used to compute the mass terms corresponding to
the in-plane displacements. The rotary inertias were also included in the consistent mass
matrix, unlike the lumped mass matrix. In the case of the thermal analysis, the temperature
136
variation was assumed to vary linearly along the thickness of the element and the variation
in the plane of the element was assumed to be linear in the standard area coordinates. The
geometrically nonlinear analysis was performed using an updated Lagrangian formulation
employing Green strain and PK2 stress measures. A linear displacement field was used for the
transverse displacement in order to compute the derivatives of the transverse displacement
that are required to compute the geometric stiffness or the initial stress matrix.
Several numerical examples were solved to demonstrate the accuracy of the formulation
for both small and large rotation analysis of laminated plates and shells. The results were
compared with those available in the existing literature and those obtained using the STRI3,
S4R5 and S4R elements of the commercial finite element package ABAQUS. The present
results agreed well with those available in the existing literature, wherever applicable,
for both linear and nonlinear analysis. In the case of nonlinear analysis, the agreement
between the present results and S4R was better than that with STRI3 or S4R5. The STRI3
element showed poor performance and some convergence problems, and both STRI3 and
S4R exhibited overly soft behavior for some nonlinear problems. The present element or
S4R did not not show such behavior. The present flat shell formulation which is based on
the assumption that the incremental rotations are moderate, appeared to be more reliable
and required fewer load steps compared to STRI3 which is based on a large rotation theory.
It is not clear why a large number of load steps are required in the case of STRI3 in spite
of the fact that the formulation is based on the large rotation theory. The reason for the
better performance of the present flat shell element could be attributed to the use of a better
membrane representation than that used in the case of STRI3.
A large number of elements were required to obtain convergence towards the exact
solution or finite element results available in the existing literature. When a flat shell
element is used for the analysis of general shells, a finer mesh is mandatory in order to
obtain a closer approximation to the curved structure. In the case of problems involving
large rotations, more steps were required for the analysis using the present element than
required by the S4R element or elements in the existing literature based on a large rotation
theory. Though a finer mesh and large number of load steps are required, the analysis is not
expected to be computationally expensive, due to the extreme simplicity of the formulation.
A comparison of the CPU time required for the analysis using the present element and those
of ABAQUS (STRI3, S4R and S4R5) has not been presented. This is due to the fact that
the present results were obtained using a code entirely developed by the author, whereas
the code used in the commercial software ABAQUS must have been well tuned for optimal
performance. However, comparing the number of load steps required by the present element
and STRI3, it can be concluded that the analysis using the present flat shell element would
certainly be faster than that using the STRI3 element, provided a similar computational
environment is used for both. The computational time can be drastically reduced by taking
proper care in developing the code like avoiding redundant computations, using data stored
in contiguous memory locations, using an efficient solver, etc. The computational time can
be further reduced by using a parallel code. The entire process of generation of element
stiffness matrices, assembly of the global stiffness matrix and the solution of the system of
equations can be performed in parallell on several processors, thereby reducing the overall
time required for the computation.
The present formulation does not suffer from in-plane rotational singularity, spurious zero
energy modes or shear locking and hence the formulation is less complicated compared to the
flat shell formulations which suffer these problems, since additional efforts are required to
alleviate such problems, often making the formulation complex. The simplicity and reliability
of the present formulation, combined with the availability of high speed parallel computers,
renders it a promising candidate for economic and yet accurate analysis of large flexible thin
structures. The present element was employed for two main applications involving large
flexible structures.
The first application is the control of thermal deformations of a spherical mirror segment.
The feasibility of controlling the surface distortions of the mirror due to arbitrary thermal
fields, using discrete and distributed actuators, was studied. This kind of study was required
for the design of a multi-segmented primary mirror of a next generation space telescope.
To determine the effectiveness of the actuators in controlling the thermal deformations of
the mirror segment, a comparative study was conducted using 2 different models of the
mirror-actuator system: (1) the mirror mounted on kinematic supports and controlled by
piezoelectric strips bonded to the rear surface of the mirror and (2) the mirror mounted
on force actuators which are used to support the mirror as well as to control the surface
deformations of the mirror. The mirror was modeled using the flat shell elements, the
piezoelectric strips were also modeled using the flat shell elements, but as a separate layer of
isotropic material and the force actuator was modeled using a beam element. Performance of
evenly distributed strips with that of strips placed at near optimal locations obtained using
heuristic integer programming was also compared. Since the force actuators are used for
both support and control, the problem does not remain the same by changing the locations
of the actuators. So no attempt was made to determine the optimal locations of the force
actuators. Results were presented for surface distortions of the mirror segment due to various
temperature distributions given in the form of linear combinations of the first few Zernike
polynomials. Both the force actuators and the piezoelectric strips were found to be equally
effective in controlling the surface deformations of the mirror. The performance of a set of
evenly placed piezoelectric strips was found to be quite compatable to that of strips placed
at near-optimal locations.
A major drawback of the force actuators is the increase in the overall weight of the
system, which is undesirable for space applications. On the other hand, the piezoelectric
strips are very lightweight and hence a large number of such strips can be used to control the
surface distortions of the mirror, without imposing a weight penalty. In the actual mirror,
the piezoelectric strips cannot be relocated as they will be bonded to the rear surface of
the mirror. Moreover, the optimal locations determined for one type of disturbance may
not be optimal for other disturbances. Correcting the mirror deformations using a set of
strips with predetermined locations thus seems to be more practical than trying to find the
optimal locations. Controlling the mirror deformations using a large number of evenly placed
piezoelectric strips seems to be a viable solution as such a configuration can impart control
to almost every part of the mirror and hence can be used to control arbitrary deformations.
In summary, the piezoelectric strips appear to be promising candidates for static shape
control of flexible structures in space, opening several avenues for further research in order
to determine the feasibility of using a large number of such strips for practical problems.
The second application for which the flat shell element was intended to be used is
the analysis of an inflatable structure. Such a structure is being considered by the US
Army to be used as a transportable tent to house Army vehicles and personnel. The tent
structure is made up of membranes suported by arches stiffened by internal pressure. The
updated Lagrangian formulation of the flat shell element has been developed primarily for
the nonlinear analysis of the tent structure since such a structure is expected to undergo large
deformations and rotations under the action of environmental loads like the wind/pressure
and snow loads. Since the direction of the pressure load is assumed to be normal to the
current configuration of the structure, its direction changes as the structure undergoes
deformation. This is called the follower action. As a result of the follower action, the
pressure load is a function of the displacements and contributes to the tangent stiffness
matrix in the case of geometrically nonlinear analysis. The thermal load also contributes
to the system tengent stiffness matrix. In the case of the thermal load, this contribution is
similar to the initial stress matrix and hence no additional effort is required to compute this
contribution. In the case of the pressure load, this contribution called the pressure stiffness
is in general unsymmetric and its determination is not straightforward. In the present
research, the deformation dependent pressure load and the pressure stiffness matrix were
systematically derived using the principle of virtual work. The contribution of the rotational
degrees of freedom was neglected in determining the pressure load and the pressure stiffness
matrix. Since most of the deformation of the element is comprised of rigid body motions,
a linear displacement field was used to compute the pressure load and the pressure stiffness
matrix. The updated Lagrangian formulation of the flat shell element, including the follower
effects as mentioned above was validated using standard examples in the existing literature
on deformation dependent pressure loads. Before attempting to analyze the tent structure
using the present element, analysis of a single circular arch under nonuniform pressure loads
was performed. The results were compared with those obtained using the STRI3 and S4R
elements of ABAQUS. The present results showed excellent agreement with those of S4R,
whereas the STRI3 did not agree well with the present results or those obtained using S4R.
After having thoroughly validated the present flat shell formulation for a wide variety of
problems and gaining confidence in using the element to obtain the nonlinear response of
large thin flexible structures, the element is currently being used for nonlinear analysis of the
tent structure under wind and snow loads. The wind load is being modeled as a non-uniform
pressure load and the snow load as lumped concentrated loads.
Bibliography
[1] Meek, J. L. and Tan, H. S., “A Faceted Shell Element With Loof Nodes”. International
Journal for Numerical Methods in Engineering, Vol. 23, pp. 49-67, 1986.
[2] Yang, T. Y., “Finite Element Structural Analysis”, Prentice-Hall, 1986.
[3] Allman, D. J., “A Compatible Triangular Element Including Vertex Rotations For Plane
Elasticity Analysis”, Computers and Structures, Vol. 19, pp. 1-8, 1984.
[4] Allman, D. J., “Evaluation of the Constant Strain Triangle with Drilling Rotations”,
International Journal for Numerical Methods in Engineering, Vol. 26, pp. 2645-2655,
1988.
[5] Cook, R. D., “On The Allman Triangle and Related Quadrilateral Element”, Computers
and Structures, Vol. 22, pp. 1065-1067, 1986.
[6] Bergan, P. G. and Fellipa, C. A. “A Triangular Membrane Element With Rotational

Degrees of Freedom”, Computer Methods in Applied Mechanics and Engineering, Vol.
50, pp. 24-69, 1984.
[7] Bergan, P. G. and Nygard, M. K.,“Finite Elements With Increased Freedom in Choosing
Shape Functions”, International Journal for Numerical Methods in Engineering, Vol. 20,
pp. 643-664, 1985.
[8] Zienkiewicz, O. C. and Taylor, R. L., “The Finite Element Method”, 4th edition, Vol.
1, McGraw-Hill, 1989.
[9] Bazeley, G. P., Cheung, Y. K., Irons, B. M. and Zienkiewicz, O. C., “Triangular
Elements in Plate Bending, Conforming and Nonconforming Solutions”, Proc. 1st Conf.
141
BIBLIOGRAPHY 142
on Matrix Methods in Structural Mechanics, pp. 547-576, AFFDLTR-CC-80, Wright-

Patterson AF Base, Ohio, 1966.
[10] Carpenter, N., Stolarski, H. and Belytschko, T., “ A Flat Triangular Shell Element With
Improved Membrane Interpolation”, Communications in Applied Numerical Methods,
Vol. 1, pp. 161-168, 1985.
[11] Fish, J. and Belytschko, T., “Stabilized Rapidly Convergent 18-Degrees of Freedom Flat
Shell Triangular Element”, International Journal for Numerical Methods in Engineering,
Vol. 33, pp. 149-162, 1992.
[12] Cook, R. D., “Some Options For Plane Triangular Elements With Rotational Degrees
of Freedom”, Finite Elements in Analysis and Design, Vol. 6, pp. 245-249, 1990.
[13] Zienkiewicz, O. C., “The Finite Element Method”, 3rd edition, McGraw-Hill, London,
1977.
[14] MacNeal, R. H. and Harder, R. L., “Refined Four-Noded Membrane Element with
Rotational Degrees of Freedom”, Computers and Structures, Vol. 28, pp. 75-84, 1988.
[15] Cook, R. D., “Modified Formulations for Nine-D.O.F. Plane Triangles That Include
Vertex Rotations”, International Journal for Numerical Methods in Engineering, Vol.
31, pp. 825-835, 1991.
[16] Batoz, J. L., Bathe, K. J. and Ho, L. W., “A Study of Three Noded Triangular Plate
Bending Elements”, International Journal for Numerical Methods in Engineering, Vol.
15, pp. 1771-1812, 1980.
[17] Cook, R. D., “A Plane Hybrid Element With Rotational DOF and Adjustable Stiffness”,
International Journal for Numerical Methods in Engineering, Vol. 24, pp. 1499-1508,
1987.
[18] Allman, D. J., “A Quadrilateral Finite Element Including Vertex Rotations for Plane
Elasticity Analysis”, International Journal for Numerical Methods in Engineering, Vol.
26, pp. 717-730, 1987.
BIBLIOGRAPHY 143
[19] Yunus, S. M., Saigal, S. and Cook, R. D., “On Improved Hybrid Finite Element
with Rotational Degrees of Freedom”, International Journal for Numerical Methods
in Engineering, Vol. 28, pp. 785-800, 1989.
[20] Cazzani, A. and Atluri, S. N., “Four-noded Mixed Elements Using Unsymmetric Stresses
for Linear Analysis of Membranes”, Computational Mechanics, Vol. 11, pp. 229-251,
1993.
[21] Hughes, T. R. J. and Brezzi, F., “On Drilling Degrees of Freedom”, Computer Methods
in Applied Mechanics and Engineering, Vol. 72, pp. 105-121, 1989.
[22] Ibrahimbegovic, A., “A Novel Membrane Finite Element With an Enhanced

Displacement Interpolation”, Finite Elements in Analysis and Design, Vol. 7, pp. 167-
179, 1990.
[23] Iura, M and Atluri, S. N, “Formulation of a Membrane Finite Element With Drilling
Degrees of Freedom”, Computational Mechanics, Vol. 9, pp. 417-428, 1992.
[24] Alvin, K., Horacio M. de la Fuente, Hhaugen. B. and Fellipa, C. A., “ Membrane
Triangles With Corner Drilling Freedoms - I. EFF Element”, Finite Elements in Analysis
and Design, Vol. 12, pp. 163-187, 1992.
[25] Fellipa, C. A. and Militello, C.,“ Membrane Triangles With Corner Drilling Freedoms
- II. ANDES Element”, Finite Elements in Analysis and Design, Vol. 12, pp. 189-201,
1992.
[26] Fellipa, C. A. and Alexander, S., “ Membrane Triangles With Corner Drilling Freedoms
- III. Implementation and Performance Evaluation”,Finite Elements in Analysis and
Design, Vol. 12, pp. 203-239, 1992.
[27] Morley, L. S. D., The Constant-Moment Plate Bending Element”, Journal of Strain
Analysis, Vol. 6, pp. 20-24, 1971.
[28] Hrabok, M. M. and Hrudey, T. M.,“A Review and Catalogue of Plate Bending Finite
Elements“, Computers and Structures, Vol. 19, pp. 479-495, 1984.
BIBLIOGRAPHY 144
[29] Jeyachandrabose, C. and Kirkhope, J.,“An Alternative Explicit Formulation for

the DKT Plate Bending Element“, International Journal for Numerical Methods in
Engineering, Vol. 21, pp. 1289-1293, 1985.
[30] Jeyachadrabose, C. and Kirkhope, J.,“Construction of New Efficient Three-node

Triangular Thin Plate Bending Elements”, Computers and Structures, Vol. 23, pp. 587-
603, 1986.
[31] Tessler, A., and Hughes, T. J. R., “A Three-node Mindlin Plate Element With Improved
Transverse Shear”, Computer Methods in Applied Mechanics and Engineering, Vol. 50,
pp. 71-101, 1985.
[32] Meek, J. L. and Tan, H. S., “A Discrete Kirchhoff PLate Bending Element with Loof
Nodes”, Computers and Structures, Vol. 21, pp. 1197-1212, 1985.
[33] Gopalacharyulu, S., “A Higher Order Conforming Rectangular Plate Element”,

International Journal for Numerical Methods in Engineering, Vol. 6, pp. 305-308, 1973.
[34] Irons, B. M., “Comment on a Higher Order Conforming Rectangular Plate Element
by S. Gopalacharyulu”, International Journal for Numerical Methods in Engineering,
Vol. 6, pp. 308-309,1973.
[35] Dhatt, G., Marcotte, L., and Matte, Y., “A New Triangular Discrete Kirchhoff Plate
Shell Element”, International Journal for Numerical Methods in Engineering, Vol. 23,
pp. 453-470, 1986.
[36] Felippa, C. A., and Bergan, P. G., “A Triangular Bending Element Based on
Energy Orthogonal Free Formulation”, Computer Methods in Applied Mechanics and
[37] Specht, B.,“Modified Shape Functions for the Three Node Plate Bending Element
Passing the Patch Test”, International Journal for Numerical Methods in Engineering,
Vol. 26, pp. 705-715, 1988.
[38] Zienkiewicz, O. C., Taylor, R. L., Papadopoulos, P. and Onate, E.,“ Plate Bending
Elements With Discrete Constraints: New Triangular Elements”, Computers and
Structures, Vol. 35, pp. 505-522, 1990.
BIBLIOGRAPHY 145
[39] Bathe, K. J., Ramm, E., and Wilson, E. L., “Finite Element Formulations for Large
Deformation Dynamic Analysis”, International Journal for Numerical Methods in
Engineering, Vol. 9,pp. 353-386, 1975.
[40] Argyris, J. H., Dunne, P. C., Malejannakis, G. A., and Schelkle, E., “A Simple
Triangular Facet Shell Element With Applications to Linear and Non-linear Equilibrium
and Elastic Stability Problems”, Computer Methods in Applied Mechanics and
[41] Horrigmoe, G., and Bergan, P. G., “Nonlinear Analysis of Free-form Shells by Flat
Finite Elements”, Computer Methods in Applied Mechanics and Engineering, Vol. 16,
pp. 11-35, 1978.
[42] Chen, K. K., “A Triangular Plate Finite Element For Large-Displacement Elastic-Plastic
Analysis of Automobile Structural Components”, Computers and Structures, Vol. 10,
pp. 203-215, 1979.
[43] Peric, D. and Owen, D. R. J., “The Morley Thin Shell Finite Element For
Large Deformations Problems: Simplicity Versus Sophistication”, Proceedings of the
International Conference on Nuclear Engineering Computations, pp. 121-142, Swansea,
Pineridge Press, 1991.
[44] Bathe, K. J., and Ho, L. W., “A Simple and Effective Element for Analysis of General
Shell Structures”, Computers and Structures, Vol. 13, pp. 673-681, 1981.
[45] Hsiao, K. M., “Nonlinear Analysis of General Shell Structures by Flat Triangular Shell
Element”, Computers and Structures, Vol. 25, pp. 665-675, 1987.
[46] Peng, X., and Crisfield, M. A., “A Consistent Co-rotational Formulation for Shells Using
the Constant Stress/Constant Moment Triangle”, International Journal for Numerical
Methods in Engineering, Vol. 35, pp. 1829-1847, 1992
[47] Meek, J. L., and Tan, H. S., “Instability Analysis of Thin Plates and Arbitrary Shells
Using a Faceted Shell Element With Loof Nodes”, Computer Methods in Applied
Mechanics and Engineering, Vol. 57, pp. 143-170, 1986.
BIBLIOGRAPHY 146
[48] Poulsen, P. N, and Damkilde, L., “A Flat Triangular Shell Element With Loof
Nodes”,International Journal for Numerical Methods in Engineering, Vol. 39, pp. 3867-
3887,1996.
[49] Fafard, M., Dhatt, G., and Batoz, J. L., “A New Discrete Kirchhoff Plate/Shell Element
With Updated Procedures”, Computers and Structures, Vol. 31, pp. 591-606,1989.
[50] Knight Jr, N. F., “The Raasch Challenge for Shell Elements”, AIAA Journal, Vol. 35,
pp. 375-381,1997.
[51] Madenci, E., and Barut, A., “A Free Formulation-Based Flat Shell Element for
Nonlinear Analysis of Thin Composite Structures, International Journal for Numerical
Methods in Engineering, Vol. 37, pp. 3825-3842, 1994.
[52] Oral, S., and Barut, A., “A Shear-Flexible Facet Shell Element For Large Deflection
and Instability Analysis”, Computer Methods in Applied Mechanics and Engineering,
Vol. 93, pp. 415-431, 1991.
[53] Simo, J. C., and Fox., D. D., “On a Stress Resultant Geometrically Exact Shell Model.
Part I. Formulation and Optimal Parametrization”, Computer Methods in Applied
[54] Parisch, H., “An Investigation of a Finite Rotation Four Node Assumed Strain
Shell Element”, International Journal for Numerical Methods in Engineering, Vol. 21,
pp. 127-150, 1991.
[55] Ibrahimbegovic, A., “Stress Resultant Geometrically Nonlinear Shell Theory with
Drilling Rotations - Part I. A Consistent Formulation”, Computer Methods in Applied
[56] Surana, K. S., “Geometrically Nonlinear Formulation for the Curved Shell Elements”,
International Journal for Numerical Methods in Engineering, Vol. 19, pp. 581-615, 1983.
[57] Brank, B., Peric, D., and Damjanic, B., “On Implementation of Non-linear Four Node
Shell Element for Thin Multilayered Elastic Shells”, Computational Mechanics, Vol. 16,
pp. 341-358, 1995.
BIBLIOGRAPHY 147
[58] Onate, E., Zarate, F., and Flores, F., “A Simple Triangular Element for Thick and Thin
Plate and Shell Analysis”, International Journal for Numerical Methods in Engineering,
Vol. 37, pp. 2569-2582, 1994.
[59] Chen, H. C., “Evaluation of Allman Triangular Membrane Element Used in General
Shell Analysis”, Computers and Structures, Vol. 43, pp. 881-887, 1992.
[60] Ertas, A., Krafcik, J. T., and Ekwaro-Osire, S., “Performance of an Anisotropic
Allman/DKT 3-Node Thin Triangular Flat Shell Element”, Composites Engineering,
Vol. 2, pp. 269-280, 1992.
[61] Kapania, R. K., and Mohan, P., “Static, Free Vibration and Thermal Analysis of
Composite Plates and Shells Using a Flat Triangular Shell Element”, Computational
Mechanics, Vol. 17, pp. 343-357, 1996.
[62] Mohan, P. and Kapania, R. K., “Updated Lagrangian Formulation of a Flat Triangular
Element for Thin Laminated Shells”, accepted for publication in the AIAA Journal.
[63] Mohan, P. and Kapania, R. K., “Analysis of General Shell Under Deformation
Dependent Pressure Loads Using a Flat Triangular Shell Element”, submitted for
possible presentation at the 39th SDM conference.
[64] Kapania, R. K., Mohan, P. and Jakubowski, A., “Control of Thermal Deformation of a
Spherical Mirror Segment”, accepted for publication in the AIAA Journal of Spacecraft
and Rockets, also presented at the 6th AIAA/NASA/USAF Multidisciplinary Analysis
and Optimization Symposium, Bellevue, WA, 1996, (AIAA paper 96-4154).
[65] Jakubowski, A. K., Mohan, P., Kapania, R. K, Crissafulli, P., and Hammerand, D., “8-
m UV/Visible/IR Space Telescope,Proceedings of SPIE-The International Society for
Optical Engineering, Vol. 2478, pp. 20-34, 1995.
[66] Hibbitt, H. D., “Some Follower Forces and Load Stiffness”, International Journal for
Numerical Methods in Engineering, Vol. 14, pp. 937-941, 1979.
[67] Loganathan, K., Chang, S. C. and Abel, J. F., “Finite Element Representation and
Pressure Stiffness in Shell Stability Analysis”, International Journal for Numerical
Methods in Engineering, Vol. 14, pp. 1413-1429, 1979.
BIBLIOGRAPHY 148
[68] Schweizerhof, K. and Ramm, E., “Displacement Dependent Pressure Loads in Nonlinear
Finite Element Analysis”, Computers and Structures, Vol. 18, pp. 1099-1114, 1984.
[69] Yuan, K. Y., and Liang, C. C., “Nonlinear Analysis of an Axisymmetric Shell Using
Three Noded Degenerated Isoparametric Shell Elements”, Computers and Structures,
Vol. 32, pp. 1225-1239, 1989.
[70] Iwata, K., Tsukimori, K. and Kubo, F., “Buckling Analysis of Shell Structures Under
Pressure Loads Using Symmetric Load Stiffness Matrix”, Proceedings of the ASME,
Pressure Vessels and Piping Conference, Vol. 175, pp.105-109, 1989.
[71] Malvern, L. E., Introduction to the Mechanics of a Continuous Medium, Prentice-Hall

Inc., 1969.
[72] Gruttmann, F. and Taylor, R. L., “Theory and Finite Element Formulation of
Rubberlike Membrane Shells Using Principal Stretches”, International Journal for
Numerical Methods in Engineering, Vol. 35, pp. 1111-1126, 1992.
[73] Christensen, E. R.,“Effect of Pressure Stiffness on the Dynamics of Solid Rocket

Motors”, Journal of Guidance, Control and Dynamics, Vol. 17, pp. 63-68, 1994.
[74] Subramanian, G. and Bose, C.J.,“Convenient Generation of Stiffness Matrices for the
Family of Plane Triangular Elements”, Computers and Structures, Vol. 15, pp. 85-89,
1982.
[75] Jeyachandrabose, C. and Kirkhope, J., “An Alternative Explicit Formulation for
the DKT Plate Bending Element”, International Journal for Numerical Methods in
Engineering, Vol. 21, pp. 1289-1293. 1985
[76] Kleiber, M., Incremental Finite Element Modeling in Non-linear Solid Mechanics,
J. Wiley, New York, 1989.
[77] Zienkiewicz, O. C. and Taylor, R. L., The Finite Element Method, 4th edition, Vol. 2,
McGraw-Hill, NY, 1989.
[78] Crisfield, M. A., “A Fast Incremental/Iterative Solution Procedure that Handles Snap-
Through”, Computers and Structures, Vol. 13, 1981, pp. 55-62.
BIBLIOGRAPHY 149
[79] Bathe, K. J., Finite Element Procedures in Engineering Analysis, Prentice Hall, Inc.,
1982.
[80] Bathe, K. J., Ramm, E. and Wilson, E. L., “Finite Element Formulations for Large
Deformation Dynamic Analysis”, International Journal for Numerical Methods in
Engineering, Vol. 9, 1975, pp. 353-386.
[81] Phan, N. D. and Reddy, J. N., “Analysis of Laminated Composite Plates Using A Higher
Order Shear Deformation Theory”, International Journal for Numerical Methods in
Engineering, vol. 21, pp. 2201-2219, 1985.
[82] Whitney, J. M. and Leissa, A. W., “Analysis of Anisotropic Heterogeneous Plates”,

Journal of Applied Mechanics, Vol. 36, pp. 261-266, 1969.
[83] Macneal, R. H. and Harder, R. L., “A Proposed Standard Set of Problems to Test Finite
Element Accuracy”, Finite Elements in Analysis and Design, Vol. 1, pp. 3-20, 1985.
[84] Reddy, J. N., “Exact Solutions of Moderately Thick Laminated She;;”, Journal of
Engineering Mechanics Division of the ASCE, Vol. 110, pp. 794-807, 1984.
[85] Kapania, R. K. and Singhvi, S., “Free Vibration Analysis of Generally Laminated
Tapered Skew Plates : Enhancing Analysis Techniques For Composite Materials” ASME
NDE-Vol. 10, pp. 159-168, 1991.
[86] Wu, C. H ; Tauchert, T. R. (1980): Thermoelastic Analysis of Laminated Plates. 1:

Symmetric Specially Orthotropic Laminates. J. Thermal Stresses. 3, 247-259
[87] Wu, C. H and Tauchert, T. R, “Thermoelastic Analysis of Laminated Plates. 2:

Antisymmetric Cross-ply and Angle-ply Laminates”, Journal of Thermal Stresses vol.
3, pp. 365-378, 1980.
[88] Buechter, N. and Ramm, E., “Shell Theory Versus Degeneration - A Comparision in
Large Rotation Finite Element Analysis”, International Journal for Numerical Methods
in Engineering, Vol. 34, 1992, pp. 39-59.
[89] Meyers, C. A. and Hyer, M. W., “Thermally-Induced, Geometrically Nonlinear

Response of Symmetrically Laminated Composite Plates”, Composites Engineering, Vol.
2, 1992, pp. 3-20.
BIBLIOGRAPHY 150
[90] Seide, P. and Jamjoon, T. M. M., “Large Deformation of Circular Rings Under
Nonuniform Pressure”, Journal of Applied Mechanics, Vol. 41, 1974, pp. 192-196.
[91] Robertson, H. J., Crane, R. and Hemstreet, H. S., “Active Optics System for Spaceborne
Telescopes”, NASA CR − 66297, Oct. 1966.
[92] Robertson, H. J., “Development of an Active Optics Concept Using a Thin Deformable
Mirror”, NASA CR − 1593, 1970.
[93] Creedon, J. F., and Lindgren, A. G., “Control of the Optical Surface of a
Thin Deformable Primary Mirror with Application to an Orbiting Astronomical
Observatory”, Automatica, Vol. 6, pp. 643-660, 1970.
[94] Howell, W. E. and Creedon, J. F., “A Technique for Designing Active Control Systems”,
NASA T ND − 7090, January, 1973.
[95] Ostroff, J. A., “Evaluation of Control Laws and Actuator Locations for Control Systems
Applicable to Deformable Astronomical Telescope Mirrors”, NASA T ND − 7276,
October 1973.
[96] Scott, R. M., “New Technique for Controlling Optical Mirror Shapes”, Optical
[97] Bushnell, D., “Control of Surface Configurations of Nonuniformly Heated Shells”,

AIAA Journal, Vol.17, pp. 78-84, 1979.
[98] Hansen, J. G. R., Richard, R. M. and Shannon, R. R., “Deformable Primary Mirror for
a Space Telescope”, Applied Optics, Vol. 21, pp. 2620-2630, 1982.
[99] Masaki, T., Noboru, I., Keizo, M., Atsushi S., Masanori, I., Yasumasa, Y., Takeshi,
N. and Wataru, T., “Shape Control Experiments with a Functional Model for Large
Optical Reflectors”, First Joint U.S/Japan Conference on Adaptive Structures, Maui,
Hawaii, November 13-15, 1990.
[100] Kuo, C. P., “Optimal Actuator Placement on an Active Reflector Using a Modified
Simulated Annealing Technique”, First Joint U.S/Japan Conference on Adaptive
Structures, Maui, Hawaii, November 13-15, 1990.
BIBLIOGRAPHY 151
[101] Kuo, C. P., “Adaptive Optics - One Meter Deformable Composite Mirror”, AIAA
Paper, 91-0907-CP, presented at the 32nd AIAA Structures, Structural Dynamics and
Materials Conference, Baltimore, MD, April 8-10, 1991.
[102] Kuo, C. P., “Optical Tests of an Intelligently Deformable Mirror for Space Telescope
Technology”, Proceedings of the SPIE - The International Society for Optical
Engineering, Vol. 2040 1993, pp. 631-646.
[103] Jonnalagadda, K. D., Blandford, G. E. and Tauchert, T. R., “Piezothermoelastic

Composite Plate Analysis Using First-order Shear Deformation Theory” Computers
and Structures, Vol. 51, 1994, pp. 79-89.
[104] Tzou, H. S. and Tseng, C. I., “Distributed Dynamic Identification and Controls of
Flexible Shells”, AIAA paper, 90-1089, 31st AIAA Structures, Structural Dynamics
and Materials Conference, Long Beach , CA, April 1990, pp. 2265-2273.
[105] Hughes, O. F., Ship Structures Design - A Rationally Based Computer Aided
Optimization Approach, The Society of Naval Architects and Marine Engineers, New
Jersey, 1988.
[106] Ha, S. K., Keilers, C. and Chang, F. K., “Finite Element Analysis of
Composite Structures Containing Distributed Piezoceramic Sensors and Actuators”,
AIAA Journal, Vol. 30, pp. 772-780, 1992.
[107] Pearson, E. and Stepp, L., “Response of Large Optical Mirrors to Thermal
Distributions”, SPIE Vol. 748, Structural Mechanics of Optical Systems II (1987), pp.
215-228.
[108] Haftka, R. T. and Adelman, H. M., “Selection of Actuator Locations for Static Shape
Control of Large Space Structures by Heuristic Integer Programming”, Computers and
Structures, Vol. 20, pp. 572-582, 1985.
VITAE
Mohan was born in Bangalore in India. He received his bachelors degree in Mechanical
Engineering from the Birla Institute of Technology and Science, India and his masters degree
in Aerospace Engineering from the Indian Institute of Technology, Madras, India, specializing
in computational structural mechanics. After his masters, he spent a year at IIT, Madras as
a project associate, working on computational fluid dynamics. He joined the PhD program
in Aerospace Engineering at Virginia Tech in January 1994 and for the last four years or
so has been working on developing a flat shell element. His areas of interest are the finite
element method, smart structures and inflatable structures. After his PhD he is planning to
start a challenging career in computational structural mechanics.

Mohan P Shell Dissertation

Uploaded by

Copyright:

Available Formats

Mohan P Shell Dissertation

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Mohan P Shell Dissertation

Uploaded by

Copyright:

Available Formats

DEVELOPMENT AND APPLICATIONS OF A FLAT TRIANGULAR

ELEMENT FOR THIN LAMINATED SHELLS

A DISSERTATION SUBMITTED TO THE FACULTY OF

Rakesh K. Kapania, Chairman

Raymond H. Plaut Eric R. Johnson

Efstratios Nikolaidis Owen Hughes

2 Finite Element Formulation 24

4 Shape Control of a Mirror Segment 100

5 Analysis of Inﬂatable Structures 125

3.1 Convergence of non-dimensionalized mid-deﬂection of a simply supported

4.1 Convergence of non-dimensionalized mid-deﬂection of a graphite/epoxy

1.1 Tent structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.1 Positive directions of βx and βy . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.1 Geometry of the Scordelis-Lo roof. . . . . . . . . . . . . . . . . . . . . . . . 77

4.1 Geometry and material properties of the L-frame . . . . . . . . . . . . . . . 119

5.1 Convergence of the deﬂection at point A, of a thin circular arch under a

1.2 Literature Survey

1.2.1 Triangular Membrane Elements

1.2.2 Triangular Plate Bending Elements

of this conforming element is the requirement of a 16 point numerical integration scheme.

1.2.3 Flat Triangular Shell Elements

1.3 Objective and Scope of Present Research

of large inﬂatable structures.

Figure 1.1: Tent structure

Finite Element Formulation

2.1 Membrane Formulation

un = u cos(γ12 ) + v sin(γ12) (2.7)

ut = −u sin(γ12) + v cos(γ12 ) (2.8)

cos(γ12 ) = (y2 − y1)/L12 , sin(γ12) = (x1 − x2)/L12 (2.9)

Figure 2.1: Positive directions of βx and βy

2.2 Plate Bending Formulation

u = zβx (x, y); v = zβy (x, y); w = w(x, y) (2.12)

where {κ} is the vector of curvatures given by

2.3 Updated Lagrangian Formulation

2.3.1 Static analysis

conﬁguration C1 at some time t = t1 is available and that a new conﬁguration C2 at some

δWe = δWi (2.19)

where V is the volume of the element in C1 , { } is the vector of incremental Green-Lagrange

{s} = {σ} + {Δs} (2.21)

{Δs} = [Q̄k ]{ } (2.22)

{ } = {e} + z {κ} −{ 0} (2.24)

{κ} = {βx,x βy,y βx,y + βy,x }T (2.26)

{ 0 } = {αx αy αxy }Tk ΔT (2.27)

{N} = {N1 } + {ΔN m } − {ΔN 0 } (2.31)

{M} = {M1 } + {ΔM m } −{ ΔM 0} (2.32)

The above equation can be rearranged as

{δe} = [G1 ]{δu,x δu,y δv,x δv,y δw,x δw,y }T (2.38)

δWe = {δa}T {f} (2.47)

where {q} is the element internal force vector given by

The variation of the product [G2 ]T [G1 ]T {N} is given by

2.3.2 Newton-Raphson method with load control

gt1 +Δt = gt1 + [K] δa − Δλ qref = 0 (2.55)

2.3.3 Newton-Raphson method with arc-length control for a

gt1 +Δt = gt1 + [K] δa − δλ qref = 0 (2.56)

δa = −[K]−1 gt1 +Δt + δλ [K]−1 qref = δā + δλ δat (2.57)

(a + δa)T (a + δa) = Δl2 (2.58)

a2 = 2(a + δā)T δat

where a0 is obtained from

Δli = Δli−1Nd /Ni−1 (2.63)

2.3.4 Newton-Raphson method for a thermal load